Three Hundred Years of Gravitation

Three hundred years of gravitation
PHILOSOPHI.t£
NATURAL IS
PRINCIPIA
MATHEMATICA.
Autore JS. NE Tl'T 0 N, 1rin. Coll. Cantab. Soc. Mathefeos

Proieffore LHC4jiano, & Societatis Regalis Sodali.
I M P R I M A T U R·
S. P E P Y S, Keg. Soc. P R lE S E S.
'J.lii i• 1686.
L 0 ND IN I,
Julfu Socictatis Reg.I.£ ac Typis J fe
o phi Streater. Pro{bt ap ud
plures Bibliop<>las. A11110 MDCLXXXVII.
i.
Three hundred years of gravitation
EDIT E D BY
S.W.HAWKING
Lucasian Professor of Mathematics, University of Cambridge
W.ISRAEL
Professor of Physics, University of Alberta, and
Senior Fellow, Canadian Institute for Advanced Research
Th« right tJ/tltt!

UnivrnilJ' nf Combri1/gr
to print anti .tell
ul/ manll<'f of hnoJ...t
U"Uf RfUlll('d hy
H,.,,,,. VIII;,, 1534.
Thr Universit)' has printc·d
tmtJ p11b/i.rh<'d fOlll/111/0US/)'
sinu IS84.
CAMBRIDGE UNIVERSITY PRESS
Cambridge
New York New Rochelle Melbourne Sydney

Published by the Press Syndicate of the University of Cambridge
The Pitt Building, Trumpington Street , Cambridge CB2 lRP
32 East 57th Street, New York, NY 10022, USA
10 Stamford Road , Oakleigh, Melbourne 3 166, Australia
© Cambridge University Press 1987
First published 1 987
Printed and bound in Great Britain by

Redwood Burn Limited, Trowbridge , Wiltshire
British Library cataloguing in publication data

Three hundred years of gravitation.
1. Gravitation
I. Hawking, S. W. II. Israel, W.
531' .14 QC178
Library of Congress cataloguing in publication data

Three hundred years of gravitation.
Includes bibliographies.
1. Gravitation. 2. Cosmology. 3. Astrophysics.
I. Hawking, S. W. (Stephen W.) II. Israel, W.
III. Title: 300 years of gravitation.
QC178 .T47 1987 531'.14 87-10364
ISBN 0 521 34312 7

Contents
List of contributors IX
Preface xi
1 Newton's Principia
S. W Hawking
2 Newtonianism and today's physics 5
S. Weinberg
3 Newton, quantum theory and reality 17
R. Penrose
3. 1 Newton's corpuscular-undulatory theory and reality 17
3.2 Stated reasons for rejection of wave theory 18
3.3 Newton and relativity 20
3.4 Newton's route to an undulatory--corpuscular picture 24
3.5 Quantum mechanics 25
3.6 Physical reality 26
3.7 Reality of the state vector 27
3.8 Quantum non-locality 29
3.9 Quantum mechanics and macroscopic physics 31
3.10 Linearity and time-evolution 32
3. 1 1 Quantum gravity and time-asymmetry 34
3.12 Time asymmetry of state-vector reduction 37
3.13 Reduction and the (longitudinal) graviton count 42
3.14 Non-locality in quantum geometry 45
References 48
4 Experiments on gravitation 51
A. H. Cook
4. 1 Introduction 51
4.2 Theoretical framework 52
4.3 The inverse square law 56
4.4 The weak equivalence principle 60
4.5 Preferred frames and locations 65
4.6 Additional deviations from general relativity 67
4.7 The measurement of the constant of gravitation 70

VI Contents
4.8 Conclusion 75
References 77
S Experimental gravitation from Newton's Principia to Einstein's

general relativity 80
C. M. W ill
5.1 Introduction 80
5.2 Newtonian gravitation and general relativity: commonalities and contrasts 82

5.3 Testing the fundamentals of gravitation theory 84
5.4 Testing the applications of gravitation theory 101
5.5 Beyond the tercentenary 120

Appendix: The parametrized post-Newtonian formalism 122
References 125
6 The problem of motion in Newtonian and Einsteinian gravity 128
T. Damour
6.1 Introduction 128
6.2 The N-extended-body problem in Newtonian gravity 132
6.3 The external and the internal problems of motion (Newtonian case) 134
6.4 The effacement of internal structure in the external problem 136
6.5 The effacement of external structure in the internal problem 140
6.6 Newton and the strong principle of equivalence 141
6.7 Solving the Newtonian problems of motion 144
6.8 The N-extended-body problem in Einsteinian gravity 145
6.9 Approximation methods 149
6.10 The post-Newtonian approximation methods 151
6.11 The post-Minkowskian approximation methods 156
6.12 Singular perturbation methods 160
6.13 The external and the internal problems of motion (Einsteinian case) 162
6.14 The effacement of internal structure in the external problem (Einsteinian case) 173
6.15 The problem of gravitational radiation damping and the relativistic Laplace effect 180
6.16 Conclusion 191
References 192
7 Dark stars: the evolution of an idea 199
W Israel
7.2 Early speculations (1784-1921) 201
7.3 White dwarfs: the first compact massive objects ( 1910-26) 205
7.4 The Chandrasekhar limit (1929-35) 212
7.5 Eddington's intervention (1935) 216
7.6 Neutron stars and gravitational collapse (1934-59) 223
7.7 The Schwarzschild 'singularity' (1916-66) 231
7.8 Quasars and relativistic astrophysics (19 51-72) 239
7.9 Non-spherical collapse: from frozen star to black hole (1964-71) 248
7.10 Towards the quantum era: the thermodynamics of black holes (1970-74) 261
References
8
266
Astrophysical black holes 277
R. D. Blandford
8 .2 Black holes in astrophysics 28 0
Contents vii
8.3 Observational evidence for black holes 297

8.4 Future prospects 321
References 323
9 Gravitational radiation 330
K. S. Thorne
9.2 The physical and mathematical description of a gravitational wave 338
9.3 The generation and propagation of gravitational waves 345
9.4 Astrophysical sources of gravitational waves 364
9.5 Detection of gravitational waves 400
9.6 Conclusion 445
References 446
10 The emergence of structure in the universe: galaxy formation and

dark matter 459
M. J. Rees
10.2 The constituents of the universe: 'dark' matter and 'luminous' matter 463
10.3 Large-scale structure and isotropy 474
10.4 Formation of protogalaxies 478
10.5 A flat (00 = 1) universe 488
References 496
11 Gravitational interactions of cosmic strings 499

A. Vilenkin
1 1.2 Some properties of strings 500
11.3 Nambu action 501
1 1.4 Gauge conditions 503
11.5 Oscillating loops 504
1 1.6 Waves on infinite strings 506
11.7 Gravitational field of a straight string 507
1 1.8 Gravitational field of a straight string (continued) 509
11.9 Physics in conical space 511
11.10 Other metrics with straight strings 513
11.11 Gravitational field of oscillating loops 514
1 1. 12 Gravitational radiation from loops 5 17
11.13 Gravitational rocket effect 519
1 1.14 Gravitational field of waves on infinite strings 521

References 522
12 Inflationary cosmology 524
S. K. Blau and A. H. Guth

12.2 Summary of the standard cosmological model 526
12.3 Problems of the standard cosmological model 530
12.4 The original inflationary universe 539
12.5 Successes of the original inflationary model 547
12.6 Problems of the original inflationary model 550
12.7 The new inflationary universe 555

12.8 Density perturbations in the new inflationary universe 561
viii Contents
12.9 Quantum theory of the new inflationary universe phase transition 570
12.10 Inflation in the minimal SU(5) grand unified theory 579
12.11 False vacuum bubbles and child universes 59 0
12.12 Conclusion 594
Appendix 596
References 597
13 Inflation and quantum cosmology 604
A. Linde
13.2 Chaotic inflation 607
13.3 Inflation and the wave function of the universe 610
13.4 Quantum fluctuations in the inflationary universe 612
13.5 Eternal chaotic inflation 61 8
13.6 Global structure of the inflationary universe and the anthropic principle 621
13.7 Conclusions 628
References 628
14 Quantum cosmology 631
S. W. Hawking
14.2 The quantum state of the universe 633
14.3 The density matrix 636
14.4 The Wheeler-De Witt equation 638
14.5 Minisuperspace 640
14.6 Beyond mini�u perspace 645
14.7 The direction of time 647
14.8 The origin and fate of the universe 649
References 651
15 Superstring unification 652
J. H. Schwarz
15.2 Classification of string theories 654
15.3 Feynman diagrams 656
15.4 String field theory 659
15.5 Anomalies 666
15.6 Compactification 670
15.7 Remaining problems and conclusions 672
References 673
16 Covariant description of canonical formalism in geometrical

theories
E.
676
C. Crnkowic and Witten
16.1 Yang-Mills theory 680
16. 2 General relativity 682
References 684
Contributors
R . Blandford
Department of Theoretical Astrophysics , California Institute of Technology, Pasadena,
California 9 1 125, USA
S. K. Blau
Department of Physics, University of Texas, Austin, Texas 787 12, U SA
A. H. Cook
The Master's Lodge, Selwyn College, Cambridge
C. Crnkovic
Department of Physics, Joseph Henry Laboratory, Princeton, New Jersey 08544, USA
T. Damour
Department of Fundamental Astrophysics, Observatoire de Paris, 5 Place Janssen, 92 195
Meudon , France
A. Guth
Department of Physics, Massachusetts Institute of Technology, Cambridge,
Massachusetts 02 139, USA
S. W Hawking
Department of Applied Mathematics and Theoretical Physics, Silver Street,
Cambridge CB3 9EW, UK
W Israel
Avadh Bhatia Physics Laboratory, University of Alberta, Edmonton T6G 211, Canada
A. Linde
P. N. Lebedev Institute of Physics , Academy of Sciences of U SSR, Leninski Prospect 53,
1 17924 Moscow, USSR
x Contributors
R. Penrose
Mathematical Institute, 24-29 St Giles, Oxford OX 1 3LB, UK
M. J. Rees
Institute of Astronomy, Madingley Road, Cambridge CB3 OHA, UK
J. H. Schwarz
Department of Theoretical Astrophysics , California I nstitute of Technology, Pasadena,
K. S. Thorne
Department of Theoretical Astrophysics, California I nstitute of Technology, Pasadena,
A. Vilenkin
Department of Physics and Astronomy, Tufts University, Medford ,
Massachusetts 02 155, USA
S. Weinberg
Department of Physics, University of Texas at Austin, Texas 787 12, USA
C. M. Will
McDonnell Centre for Space Science, Department of Physics, Washington University, St
Louis, Missouri 63 130, USA
E. Witten
Department of Physics, Joseph Hen ry Laboratory, P rinceton , New Jersey 08544, USA
Preface
On 5 July 1687 Edmond Halley wrote from London to Isaac Newton in

Cambridge concerning his book, the Philosophiae Naturalis Principia
Mathematica, which, in view of the Royal Society's straitened circumstances
at the time, Halley had undertaken to publish at his own expense:
Honoured Sr
I have at length brought your Book to an end, and hope it will please
you. The last errata came just in time to be inserted. I will present from
you the books you desire to the R. Society, M r Boyle, M r Pagit, Mr
Flamsteed and if there be any elce in town that you design to gratifie
that way; and I have sent you to bestow on your friends in the
University 20 Copies, which I entreat you to accept.
Three centuries later, the heavens themselves appear to have joined in the
commemoration of this historic moment. At the time of writing, neutrino, X
ray and optical astronomers are sorting out the first messages from
Supernova 1987a, at once a tribute to the birth of gravitational theory and a
signpost to its future.
We had the same dual purpose in mind when soliciting contributions to
this volume. We wanted to mark the 300th anniversary of this important
occasion with suitably reflective essays. But gravitational theory is very far
from being a closed chapter of scientific history, and most of the articles
review current developments and future prospects. They are self-contained,
but designed also to supplement and update the articles in a book we edited
previously to mark the centenary of Albert Einstein's birth. t
t General Relativity: An Einstein Centenary Survey, ed . S . W. Hawking and W . Israel.

Cambridge University Press : Cambridge 1979.
Xll Preface
The intervening eight years have witnessed dramatic progress on both the
observational and theoretical fronts. The earlier volume was already in press
when Joseph Taylor announced measurements of the orbital speed-up of the
binary pulsar which confirmed the value predicted from Einstein's formula
for the loss of energy by gravitational radiation. In 1979, this spectacular
vindication could be noted only as a last-minute tabular entry in the proof
stages of Clifford Will's article. In Chapter 5 Will carries the story of this and
other experimental tests of Einstein's theory up to the present.
The flood of data from the binary pulsar has spurred a vigorous
theoretical effort finally to bring under control the two-body problem in
general relativity, long embroiled in controversy and confusion. Thibault
Damour, who has been in the vanguard of these advances, reviews their
current status in Chapter 6.
It is now a realistic prospect that, before the turn of the century, direct
observation of gravitational waves using laser interferometry, will have
become possible and even routine. In Chapter 9, Kip Thome presents a
comprehensive review of the relevant theoretical and experimental aspects.
Cygnus X- 1 , the only firm black hole candidate known in the 1970s, has
since been joined by other binary X-ray sources in our galaxy and in the
Large Magellanic Cloud. Meanwhile circumstantial evidence for
supermassive holes in the nuclei of galaxies (including our own) has
continued to build up steadily. Roger Blandford critically reviews the
current observational evidence for black holes in Chapter 8 .
Of the recent theoretical developments, one o f the most publicised and
most exciting has been superstring theory (reviewed in Chapter 1 5 by John
Schwarz, a leading pioneer in the field) , which for the first time offers the
hope of including the gravitational force within a unified and finite theory.
Another is the great upsurge of interest in the very early universe, sparked by
Alan Guth's formulation of inflationary cosmology in 198 1 . Guth's idea
was, in part, inspired by a brilliantly imaginative chapter in the Einstein
centenary volume co-authored by Robert Dicke and Jim Peebles, which
spotlighted the horizon problem of the standard big-bang cosmology as well
as 'the remarkable balance between mass density and expansion rate'.
Chapters 12-14 by Blau and Guth, Linde and Hawking survey current ideas
on inflationary and quantum cosmology.
The inflationary hypothesis makes it difficult to escape the conclusion
that perhaps 99 % of the matter in the universe must be in some invisible
form . Martin Rees in Chapter 10 discusses various possibilities for the
nature of this hypothetical dark matter and its influence on the manner in
Preface Xlll
which galaxies form. The mechanism of galaxy formation is being actively

studied , and this b ranch of cosmology has developed into a mature science
in which detailed theoretical scenarios are matched to sophisticated
observational data on the clustering of galaxies, the isotropy of the
microwave background and the abundance of the light elements.
These brief remarks must serve as a very selective and inadequate
introduction to the many notable recent advances and to the contents of this
volume.
The book is being published in association with a Newton Tercentenary
Conference (29 June to 4 July 1 987) at Trinity College, Cambridge -
Newton 's home from 166 1 until 1696. We thank the contributors for their
excellent work and the staff of Cambridge University Press for performing
their customary miracle. Our secretaries, Cheryl Billington and Mary Yiu,
as usual have been a tower of strength. We gratefully acknowledge the
financial support of Trinity College, Cambridge, the Ralph Smith
Foundation , the Canadian Institute for Advanced Research and the Natural
Sciences and Engineering Research Council of Canada.
S. W. Hawking
W. Israel
1
Newton's Principia
S .W. H AWK I N G
The Philosophiae Natura/is Principia Mathematica b y Isaac Newton, first

published in Latin in 1687, is probably the most important single work ever
published in the physical sciences. Its significance is equalled in the
biological sciences only by The Origin of Species by Charles Darwin . The
original impulse which caused Newton to write the Principia was a question
from Edmond Halley as to whether the elliptical orbits of the planets could
be accounted for on the hypothesis of an inverse square force directed
towards the Sun. This was something that Newton had worked out some
years earlier but had not published , like most of his work on mathematics
and physics. However, Halley's challenge, and the desire to refute the
suggestions of others such as Hooke and Descartes, spurred Newton to try
to w rite a proper account of this result . In order to do so he had first to
develop a Theory of Mechanics and the mathematical techniques needed
to analyse this theory. The old Aristotelian idea had been that rest was the
natural state of a body and that a body moved only if driven by some cause.
However, Newton substituted for this his First Law: 'Every body continues
in its state of rest, or of uniform motion in a straight line, unless it is acted
upon by a force. ' Newton's Second Law stated that the acceleration, or rate
of change of velocity, was proportional to the force.
These laws formalised the Theory of Mechanics that Galileo had
implicitly assumed though not explicitly stated . In the case of a force that
was constant in magnitude and direction , like the gravitational force near
the surface of the Earth, Galileo had shown that they implied that the
distance moved by a body initially at rest would be proportional to the
square of the time elapsed . However, the question of the motion of the
planets was much more difficult because the force on them varied both in
2 S. W. Hawking
magnitude and direction. Newton seems to have solved the problem

originally using what we would now call differential calculus, which he had
invented along with integral calculus. However, for some reason he did not
think that calculus, or 'the theory of fluxions', as he called it, was suitable for
publication in the Principia. H e therefore developed a new method to show
that a body that was attracted towards a fixed point by a force proportional
to the inverse square of the distance would move on an elliptical orbit and
would obey the laws that Kepler had discovered by observation. Newton's
failure to publish his theory of fluxions led Leibnitz to discover calculus
independently and to a bitter dispute about priority .
Not only did Newton explain the elliptical orbits of the planets, but he
also showed that the same force of gravity was responsible for the orbits of
the satellites of Jupiter and Saturn, the orbit of the M oon about the Earth,
the tides in the sea and the fall of an apple off a tree. In other words, Gravity
is Universal : every body in the universe attracts every other body with a
force that is proportional to its mass and inversely proportional to the
square of the distance between them. By Newton's Third Law, which states
that action and reaction are equal and opposite, every body is similarly
attracted towards every other body. It was natural to suppose that this
universality might extend to light which was first shown to travel at a finite
velocity by Roemer in 1676. Roemer observed that there were apparent
discrepancies between the timing of the orbits of the moons of Jupiter and
the predictions of Kepler's theory. These discrepancies disappeared if he
assumed thatJight travels at a finite velocity and therefore takes a variable
amount of tlme to cross the variable distance between Jupiter and the Earth.
If light travels at a finite velocity and if it is affected by gravity in the same
way as planets or cannonballs, then one can imagine a situation in which
one has a body which is so massive and compact that light cannot escape
from it but gets dragged back by gravity. The possible existence of such
objects was pointed out by Michell in 1783 and by Laplace in 1799 but not
much notice was taken at the time. Indeed, Laplace removed the suggestion
from the later editions of his book Exposition du Systeme du Monde.
However, we now believe that these objects, which we call black holes, are
the final stage of evolution of massive stars and that matter falling into them
provides the power for compact X-ray objects and the radio and optical
emission that we observe from quasars, very compact and luminous sources
that we can see even at the other side of the universe.
Another consequence of the universality of gravity that was not properly
appreciated until much later was that it was inconsistent with the prevailing
Newton's Principia 3
idea that the universe was infinite in extent and filled with a distribution of
stars with a density that was more or less constant in space and time. If the
stars were at rest with respect to each other at one time, the gravitational
attraction of each star for the others would cause them to start to fall
together and the density would go up with time. Alternatively, the stars
could be moving away from each other. In that case the density would
decrease with time. Gravity would act to slow down the recession of stars
from one another and might eventually stop the expansion of the universe
and convert it into a contraction . Newton was aware of this problem but he
argued that, in an infinite distribution of stars, the force on a star caused by
the attraction of other stars to one side of it (say, the right side) would be
almost exactly balanced by the force of attraction of the stars to the left of the
star. The net force on the star would therefore be small and the system of
stars could continue to exist at more or less constant distances from each
other. However, there is a flaw in this argument . The force of attraction
produced by the stars to the right of the given star would be infinite because,
although the force produced by each star to the right gets smaller the further
away the star is from the given star, the number of stars at a given distance
gets larger the fu rther one is from the given star. Similarly, the force of
attraction produced by the stars to the left would be infinite. When one
subtracts infinity from infinity, it is well known that one can get any answer
one wants. We now know that the universal attractive nature of gravity is
inconsistent with a static infinite universe : the universe has to be either
expanding or contracting. This was not realised, however, until in the 1920s
observations of distant galaxies revealed that the universe is expanding. It
was a great missed opportunity for theoretical physics : Newton could have
predicted the expansion of the universe.
In the Principia Newton introduced the concepts of Absolute Time and
Absolute Space. By Absolute Time he meant that there was a quantity called
time which would be measured by all properly constructed clocks,
regardless of their motion. This was an idea that nobody questioned until
the advent of the Theory of Relativity at the beginning of this century. Much
more controversial was Newton's advocacy of Absolute Space. Newton
argued that in the case of circular motion there was a preferred state of rest in
which there were no centrifugal forces. He therefore claimed that by analogy
there ought to be a preferred state of Absolute Rest with respect to linear
motion, though he admitted that it would be difficult to determine this state
of rest by observations. He suggested, as a hypothesis, that this state of rest
coincided with the centre of gravity of the solar system. In fact Newton's
4 S. W Hawking
Laws of Motion are the same in all frames of reference moving uniformly in
straight lines though they are not the same in rotating frames. Thus the laws
do not determine a state of rest and so do not support the concept of
Absolute Space, though they are not inconsistent with it. They do, however,
require the concept of Absolute Time. In other words, in Newton 's theory
one can determine whether two events at different points in space occur at
the same time but one cannot tell whether events at different times occur at
the same point of space : that depends on the frame of reference and would be
different in two frames that were moving relative to each other.
The laws presented in the Principia remained the accepted theories of
mechanics and gravity for more than two hundred years. Even today they
are the basis of nearly all practical calculations. It is only in very extreme
situations that one has to take into account the modifications introduced by
the Special and General Theories of Relativity formulated by Albert
Einstein in 1 905 and 1 9 1 5 respectively. In the Theory of Relativity one has to
abandon the concept of Absolute Time as well as that of Absolute Space.
The time measured by a clock now depends on its velocity and one can n o
longer determine whether events at different points o f space occur a t the
same time : again this depends on the motion of the frame of reference. The
philosophical implications of this change from Absolute Space and Time to
Relative Space and Time have been profound but have often not been
properly appreciated. Lenin wrote a pamphlet attacking Relativity because
he thought it was a threat to the absolute system of Hegel and M arx. It is
only in the last thirty years that the study of relativity has become
respectable in the Soviet Union. Other people in the West reacted in a
similar way. Einstein was accused of undermining moral standards by
suggesting that everything was relative. However, this attack and that of
Lenin were based on a misunderstanding of the Theory of Relativity : it is a
mathematical model of space and time. It makes no statement about how
human affairs should be conducted or organised.
Einstein is the only figure in the physical sciences with a stature that can
be compared with Newton. Newton is reported to have said : 'If I have seen
further than other men, it is because I have stood on the shoulders of giants.'
This remark is even more true of Einstein who stood on the shoulders of
Newton. Both Newton and Einstein put forward a theory of mechanics and
a theory of gravity but Einstein was able to base General Relativity on the
mathematical theory of curved . spaces that had been constructed by
Riemann while Newton had to develop his own mathematical machinery. It
is therefore appropriate to acclaim Newton as the greatest figure m
mathematical physics and the Principia is his greatest achievement.
2
Newtonianism and today's physics
S T EV E N W EI N B E R G
The written version o f my talk at the Newton tercentenary celebration in

Cambridge was not ready in time to appear in this book. Instead, I have
submitted this talk, which was originally presented at a similar celebration
at the University of Texas at Dallas, on 16 April 1986. I am grateful to the
editors for allowing me to make this substitution.
Newton was a complicated man , who lived in complicated times. It would

take the talents of a historian to say anything new and useful about Newton
himself. I am not a historian, and know better than to try. I am a physicist,
and will talk here not about Newton , but Newtonianism; that is, about
Newton's work , his achievement , as an example which over the last three
hundred years has guided the evolution of the science in which I now work ;
Newton, in other words, to use a much over-used word, as paradigm.
The central part of the Newtonian achievement I suppose is his theory of
the solar system, which had three parts, each one of which would by itself
guarantee his immortality. First, there is Newton's theory of gravity, which
explains how the distribution of all the matter in the universe determines the
gravitational force on any one particle, whether that particle is a particle
here on earth or a particle on the sun or the moon or the stars . Second, there
is a law of motion that describes how, given the forces acting on a particle,
one can calculate its motion . Third , there is a method of calculation , the
mathematical tool now known as calculus. (I gather that Newton was
apparently not at ease about calculus. He often presents his derivations in
terms of historically sanctified geometrical methods. However, calculus has
become the machinery by which subsequent generations of physicists have
learned to think about these problems.)
6 S. Weinberg
The first question that I as a physicist should address is : how well does it
work? How does Newtonianism stand up after three hundred years of
experimental testing? It stands up very well. We do now understand that
there are small corrections to the Newtonian picture of gravity and
dynamics. The most interesting and important corrections are those
provided by Einstein's general theory of relativity, but they're awfully small.
For example, in the motion of the earth around the sun, the corrections due
to general relativity are of the order of one part in a hundred million, so
small in fact that they've never been detected . We detect the effects of
relativity only for more rapidly moving bodies like the planet Mercury, and
for light itself.
However, whether the corrections are large or small doesn't seem to me to
be the point. In 19 19, when Einstein's theory began to be popular, The
Times of London proclaimed that Einstein had disproved Newton . That's
very far from the truth. Einstein's theory reduces to Newton's theory in the
limit of slowly moving bodies at large distances from each other, which is
certainly true of the outer part of the solar system, and indeed of most of the
umverse.
In fact, it is fair to say that, not only does Einstein's theory not supplant
Newton's theory , it explains Newton's theory. (I thi �k this is a point that's
not often appreciated.) In Newton's work the inverse square law appears as
a means of accounting for the observations of the solar system, particularly
Kepler's interpretation of Brahe's observations in terms of a relationship
between periods and radii of orbits. For Newton it was just a fact , learned
from experiment, that the gravitational force fell off as the inverse square of
the distance. As far as Newton was concerned, ifthe force had fallen off as the
inverse cube of the distance or the inverse 2 . 1 power of the distance that
would have been perfectly fine. There was no explanation in Newton's
theory of why it had to be an inverse square law. This explanation was finally
provided in 1 9 1 5 and 19 16, by Einstein's general theory of relativity. If you
adopt Einstein's insight that the force of gravity is due to a curvatu re in
spacetime, and follow that insight to where it leads you , you find that you
cannot without standing on your head make a theory of gravity in which the
force at large distances is anything but an inverse square law. (What the
force is at short distances is a very complicated question to which I will
return.)
Newton's theory then works very well , and it now has a rationale that it
lacked in Newton's time, provided by general relativity. We use it to follow
the motion of objects in the solar system, some of them objects that we put
N ewtonianism and today's physics 7
there ou rselves. I think nothing was more dramatic as an example of our

faith in the Newtonian theory than the Apollo project . In its early stages,
remember that astronauts were sent out in spacecraft which circled the
moon, orbited several times, were given a little boost from a rocket motor,
and then came back to the earth with essentially all the fuel gone; just relying
on the validity of Newton's laws. It was somehow very appropriate that
Major William Anders on the return from the first circumlunar voyage in
December 1968 , made the remark , 'I think Isaac Newton is doing most of
the driving right now.' (I'm sure that other talks and articles about Newton
this year will provide you with wonderful quotes about Newton , quotes
from Pope and Halley and Blake; I just wanted to drag in one quote that
they probably hadn't used.)
So Newton's theory is not challenged as a theory of the solar system.
What about Newton's theory as a model for physical theory in general?
I think the qualitative feature of Newton's theory that has come under the
most intense and controversial discussion in this century has been its
determinism. (Though Newton himself may not have been so much of a
determinist - not as much, say, as Leibniz.) Look at the picture that
Newtonianism provides: At every instant each particle in the universe
experiences a force determined by the positions of all the other particles and
thei r masses. Knowing that force at that instant you can calculate the
.
acceleration ; knowing the acceleration and velocity at one instant you can
then calculate the velocity that the particle will have in the next instant . Also,
knowing the velocity it has, you can. calculate the position it will have in the
next instant. So, the solar system, and in fact the whole universe, goes
grinding along, with its configu ration in any one instant determining what
its configuration will be in the next instant. This is a picture that is deeply
satisfying to some and frightening to others.
In this century the determinism of Newton's theory came under profound
challenge from the greatest revolution in physics since the time of Newton,
the advent of quantum piechanics in the 1920s. Quantum mechanics was a
new way of looking at nature, that was particularly appropriate for matter
on the scale of atoms and, below the scale of atoms, for atomic nuclei and
elementary particles. Quantum mechanics is the basic tool we use to
understand physics in the small. Quantum mechanics , as I think almost
everyone has heard, describes natu re not by means of a language of particles
that have definite positions and velocities and move along clearly defined
trajectories, but instead in what sounds like a much cloudier language, a
language of probabilities. We are not allowed to ask where an electron is
8 S. Weinberg
inside a hydrogen atom at any one time, and if we try to find out, the
experiment that we do breaks up the atom, and we are unable to answer the
question that we set out to answer. Because in quantum mechanics one talks
in terms of probabilities, there has grown up especially in the last decade or
so an idea, fostered by some popular books, that quantum mechanics is
somehow closer to a gentler, more mystical , view of nature than the hard,
brutal deterministic view of Newton. Here Newton is being cast as a kind of
cosmic Scrooge who denies particles any volition, whereas the developers of
quantum mechanics in the 1920s are seen as gentle, soft-spoken flower
children who are going to return physics to a universal mysticism.
Nothing could be farther from the truth. Quantum mechanics in fact
provides a completely deterministic view of the evolution of physical states.
In quantum mechanics, if you know the state at one instant, and of course if
you know the machinery of the system, you can then calculate with no lack
of precision what the state is in any subsequent instant. The indeterminacy
of quantum mechanics comes in when human beings try to make
measurements, because in making measurements we inevitably disturb the
system in ways that can't be predicted, but the system itself evolves in a
completely deterministic way.
In a sense a more interesting challenge to the determinism of Newton's
theory came more recently than quantum mechanics, in the last decade or
so, within Newtonian mechanics itself, with the discovery of the importance
of what is called chaos. Chaos has become a technical term, referring to the
practical unpredictability of systems that are often very simple. It has always
been of course understood that if you have to deal with a very complicated
system with many individual parts, such as the stock market or the weather,
it will be impossible, just bec'ause the system is so complicated, to predict
what will happen. What has been realized in recent years is that even very
simple systems exhibit behaviour which can best be called chaotic, because it
is for all practical purposes unpredictable. Let me give you an example, and,
since this is the year of Halley's comet, let me give you an example involving
three bodies: a comet, a planet (let's say Jupiter), and the sun . Imagine them
all going around in their orbits, and suppose for simplicity (what is not true
for Halley's comet) that they're all on the same plane, the plane of the
ecliptic. Also for the sake of simplicity forget their finite size, so that we don't
have to worry about whether the comet is going to run into the sun or
Jupiter is going to hit the comet. The comet goes way out very far from the
sun, and then comes back on a very eccentric orbit. Jupiter goes chugging
around the sun in its nearly circular orbit. Every once in a while the two
bodies come close together. When that happens the comet is perturbed by
the gravitational field of Jupiter and goes off in a slightly different direction.
Now, there is no question that if we know with absolute mathematical
precision the motion of the comet at one instant (and if we h�d an infinitely
capable computer) we would be able to predict it at all future instants. But
that's never the way the world works. In this situation I think you can see
that if we only know the motion of the comet to, say, a tenth of a per cent
before its close encounter with Jupiter then , since what happens when it
makes the close encounter with Jupiter depends so sensitively on exactly at
what angle it approaches the planet , its motion after its encounter with
Jupiter will be more uncertain , let's say by one per cent. (Just take these
numbers as an example.) Then on the next encounter the uncertainty with
which the orbit of the comet is known increases by another factor of ten, and
so on. Clearly no matter what kind of accuracy you have when you first
observe the comet , whether you observe the elements of its orbits to one part
in a million or one part in a hundred million or what you will , after a
hundred or so encounters with Jupiter the comet's motion will be
completely unknowable, because the system puts such stringent demands on
the accuracy of the initial data (and on the accuracy of your computations)
that as time passes eventually no conceivable precision would allow you to
predict what would happen. The system is chaotic ; determinism has lost its
point.
This is a rather obvious example, one of many that have been known for
generations. What has been realized in the last few decades is that chaos is
itself interesting ; there are universal rules of chaos. As long as you don't try
to follow a precise trajectory, but think statistically of families of trajectories,
you can describe chaotic behavior in terms that turn out to have common
features for the orbits of comets, for the concentrations of various chemicals
in chaotic chemical reactions, for convection in fluids, and so on.
Now there are international conferences on chaos, and journals devoted
to chaos, but it took a while to realize how interesting chaos is. One of the
reasons it took so long is that our solar system is not particularly chaotic.
The orbits of the planets are quite regular (a good thing for us!). Just irt the
last year evidence has come in that one of the moons of Saturn , Hyperion , a
moon a few hundred miles in diameter shaped somewhat like a large
hamburger, is tumbling in what appears to be a chaotic manner. Here
Newtonian mechanics is dealing with what is after all a fairly simple system,
the moons of Saturn, a dozen or so moons going around the planet , the
planet going around the sun . Yet Newtonian mechanics is for all practical
10 S. Weinberg
purposes incapable of calculating what will happen next in the rotation of

the moon, Hyperion. However, such chaos is very rare in the solar system.
Apparently the reason that there is so little visible chaos in the solar system
is because of a principle I suppose could be called 'survival of the stablest'.
All those bodies which happened to be formed in orbits where they would
exhibit chaotic motion have by now received kicks which have knocked
them into other orbits or out of the solar system. If you look, for example, at
a photo of the rings of Saturn; you will see beautiful bright rings, separated
by dark lanes where there are no little moonlets. These dark lanes are the
regions within the rings where the orbits would be chaotic. You don't see
chaotic orbits in the solar system because if they were chaotic they wouldn't
last very long.
This discussion of determinism has pointed up a moral . Great ideas like
Newtonian determinism are not easily overthrown , but they do get refined
as time passes. Both quantum mechanics and the theory of chaos have
forced us to sharpen our understanding of determinism , in different ways.
Neither atomic physics nor celestial mechanics are enti rely deterministic,
but nevertheless there is a sense in which underlying these phenomena there
operates a completely deterministic mechanics, whether it's a Newtonian
mechanics or quantum mechanics.
I would not say that either the broad picture of a det e rministic mechanical
universe, or the particular details of Newton's law of gravity, or Newton's
equation of motion , or even the invention of the calculus constitute the heart
of the Newtonian achievement. In preparing this, I did some soul searching
about why Newtonianism seems to me to be such an incredible watershed in
the history of our species. I think it is because with Newton's work mankind
for the first time saw the glimpse of a possibility of a comprehensive
quantitative understanding of all of nature.
Now of course Newton was not the first person to think about nature in
quantitative terms. The Hellenistic Greeks , such as Eratosthenes, Ptolemy,
and Archimedes, and then in the century before Newton physicists like
Huygens and Galileo, were masters at bringing quantitative methods to
bear on physical reality. Also Newton was not the first person who tried to
think about nature in comprehensive terms, who tried to make a unified
theory of everything. That goes back even further than the Hellenistic
Greeks, to Thales and Anaximenes, and then , closer to Newton's time,
Descartes. But the two styles had never come together before. Those who
thought comprehensively about nature from Thales to Descartes had never
confronted the necessity of carrying their comprehensive world picture
N ewtonianism and today ' s physics 11
through to a quantitative understanding of nature i n all its aspects. And

those who thought quantitatively about nature had never seen the beginning
of a hope of formulating laws that would describe everything.
"
With Newton , for the first time one saw the possibility or a quantitative
understanding of everything, not just the motion of the planets around the
sun , but of all physical phenomena. Newton showed us what could be done.
In this sense, the most important part of Newton's opus was not his theory
of planetary motion, but his theory of the motion of the moon, in book three
of the Principia. There Newton observes that on earth the force of gravity
makes objects fall with a certain acceleration, which can be described by
saying that a body allowed to drop from rest in the first second will fall
sixteen feet . The surface of the earth where we do our experiments is a certain
distance from the center of the earth, and Newton knew that you could think
of the earth as if all its mass was concentrated at the center. (This was in fact
the hardest part of the whole argument to demonstrate mathematically.)
The moon is sixty times farther from the center of the earth than we are here
on the earth's surface, so from the inverse square law the moon, if allowed to
drop from rest, in the first second should fall not sixteen feet, but an amount
less than that by a factor of sixty squared, or 3600. You can easily work out
that sixteen feet divided by 3600 is about a twentieth of an inch. So if you let
the moon drop from rest, it should fall toward the earth a twentieth of an
inch in the first second. Now of course the moon isn't at rest, it's going
around in a circular orbit, but it's really the same thing. If the moon were not
attracted to the earth it would go off in a straight line, a tangent to wherever
.
it was in its orbit at that moment. Instead of going in a straight line it curves
toward the earth ; that's what a circle does. And any one second it curves
toward the earth an amount which brings it closer to the earth than if it had
gone in a straight line in that second by an amount which is in fact just a
twentieth of an inch. The beautiful thing about this argument is that Newton
was not just relating celestial phenomena to each other. He was relating a
celestial phenomenon, the moon's orbit, to something here on earth, an
apple falling on his head in Cambridgeshire. (No, I don't know if the apple
ever fell on his head .) Newton broke down the barrier between the celestial
and terrestrial styles in physics, and in 'so doing he not only demystified the
heavens, he opened up the possibility of an understanding which would
embrace the heavens and the earth, all in one synthesis.
I don't think there have been many things in the history of man
more important. It changed the mentality of human beings, the way they
look at the universe and at their position in the universe. The historian
12 S . Weinberg
Trevor-Roper has credited Newton and the ideas that flowed from Newton
with the disappearance of the witch craze in Europe in the eighteenth
century. Historians have argued to what extent Newtonianism , the physics
of Newton and his followers, was responsible for the industrial revolution a
century later, and I don't have an opinion of my own . The general view
seems to be that the industrial revolution was made by men like Watt and
Stephenson and Edison who were not very learned in the science of their
time (or even of Newton's time), but who operated within a scientifically
oriented style that had been created by the scientific revolution of the
seventeenth century. It's not surprising that the British historian , Herbert
Butterfield, a historian both of science and of the politics of the seventeenth
century, made the remark , 'Since the rise of Christianity there is no
landmark in history that is worthy to be compared with this'.
But Newton of course only provided us with a glimpse of a really
comprehensive system of nature. Newton knew there were other forces in
nature besides gravitation , and he knew that he didn't know what they were,
but he hoped that the same kind of mathematical reasoning , the same clarity
of vision that had revealed the nature of the force of gravitation through its
role in the solar system, would reveal the nature of the other forces, and their
role governing all the phenomena of nature. In the preface to the first edition
of the Principia, which Newton wrote at Trinity College on 8 May 1686, he
said, 'I wish we could derive the rest of the phenomena of nature by the same
kind of reasoning for mechanical principles, for I am induced by many
reasons to suspect that they may all depend on certain forces.' But he didn't
know what those forces were, and it took a long time to understand what
they were. There was the understanding in the nineteenth century by organic
chemists that the chemicals of living things were not subject to separate
chemical laws, but were subject to the same chemical laws as ordinary
inorganic chemicals. There was the understanding by Darwin and Wall ace
that the growth of various species of living things on earth did not have to be
accounted for by some extra biological laws separate from the ordinary laws
of physics and chemistry, but could be accounted for by the operations of
chance on inheritable variations among organisms. There was the
realization by Maxwell that light was not something separate from the rest
of nature, but was a manifestation of oscillating electric and magnetic fields.
And so on . These were all great steps in the devel opment of a unified view
of nature, but by the beginning ?f the twentieth century we were still a long
way from understanding even the possibility of a really unified view .
Remember the famous remark , made in 1903 by the American physicist
A. A. M ichelson. In his book, Light Waves and Their Uses (as quoted by
R. S. Andersen), he said that 'The more important fundamental laws and
facts of physical science have all been discovered, and these are now so firmly
established that the possibility of their ever being supplemented in
consequence of new discoveries is exceedingly remote. ' Physicists have
laughed at this ever since. Well , of course Michelson was wrong, but I think
most of those who laugh at this remark may miss the point about where he
went wrong. A certain kind of physics was indeed coming to an end in 1903.
It was the physics of the macroscopic motions of fluids and solid bodies.
Michelson could not possibly have thought that the physicists of his time
had succeeded in explaining chemical forces, for example. To Michelson ,
chemistry and physics were two different sciences; it was not the
responsibility of physics to explain chemistry. M ichelson's mistake I would
guess was in having too unambitious a view of what would ultimately be
explained by the methods of physics. Indeed it was not until the discovery of
direct evidence that nature is composed of atoms, and the beginning of the
task of pinning down their properties - measuring the charge of the electron,
measuring the mass of the hydrogen atom - in the first few decades of the
twentieth century, that physicists really began to see that the properties of
matter - chemical forces, frictional forces, all the familiar things that happen
when you bang things together, could be accounted for in terms of physical
forces, just the way Newton had hoped .
By 1 9 1 8 , Einstein had concluded that we were well along the path to a
unified view of nature, and in a speech he gave in that year he said, 'The
supreme task of the physicist is to arrive at those universal elementary laws
from which the cosmos can be built up by pure deduction' - just Newton's
hope.
Now there's a bad name for this sort of hope : it's sometimes called
reductionism. In fact there's no question that a naive kind of reductionism
can do a great deal of harm. It is certainly not true that physics is going to
replace the other sciences. The work of the chemists will continue to be done
in terms of chemical phenomena, and chemists will not reduce all of their
efforts to solving the Schrodinger equation of quantum mechanics. It's even
more true of the work of biologists and sociologists and economists. They
will all operate within their own disciplines, because at a certain level of
complexity nature begins to exhibit phenomena that cannot be usefully
reduced to the motion of elementary particles.
We even see this within physics itself. Thermodynamics, the science of
heat, is a separate branch of physics. We do not study the behavior of hot
14 S. Weinberg
bodies when they cool by going back at every point to the motions of the
elementary particles within them. It isn't useful ; in fact if you could follow
the motion of every elementary particle in a glass of water as the water boils,
you would have an incredible amount of information about the trajectory of
every particle, and nowhere in the mountain of computer tape would you get
the impression that water was boiling.
So reductionism has its limits. Another reductionist fallacy that has to be
avoided is seeing physics as a model for the other sciences, or for human
thought in general. I think a great deal of harm has been done by those from
Herbert Spencer on who tried to see the social sciences for example as being
based on physics as a model. I think physics is a terrible model for the social
sciences. In fact it's probably a terrible model for everything except physics
itself.
Nevertheless, despite all the caveats that I've presented you with about
the dangers of reductionism , still there is a sense that among all the holes in
our understanding of nature, there is a special importance to the hole at the
bottom, that is, in our understanding of the forces which govern the
particles, out of which the atoms, out of which ordinary matter is made.
Of course we are still struggling to complete this understanding . We now
understand the·electroweak forces which are respon�ible for electricity and
magnetism and for the weak nuclear interactions. We understand the strong
nuclear forces which hold together the quarks inside the particles inside the
nucleus of the atom, and we understand some aspects of gravitation. Oddly
enough, of all the forces that we know about, the one that w.e've studied
longest, gravitation , is the one we understand least. We have a good
mathematical theory of the strong nuclear forces, known as quantum
chromodynamics, developed by about a dozen physicists in the early 1970s.
Quantum chromodynamics is a very satisfactory mathematical theory, in
which there apparently are no mathematical inconsistencies, and which
accounts as far as we can tell for all the phenomena having to do with the
strong nuclear forces, the forces that hold the nucleus of the atom together.
The electroweak forces are also well understood though with certain large
gaps in our understanding, which we hope will be closed by the next
generation of accelerators. These theories make mathematical sense, they've
been reasonably well tested by experiment , and they even have a kind of
compelling quality. They haven't been carefully adjusted to fit the data ; they
are what they are, because you can't think of anything else.
Gravity is in a very different position. We have a theory of gravity ,
Einstein's theory of general relativity, which as I said reduces to Newton's
Newtonianism and today's physics 15
theory at large distances and small velocities. This theory of gravity works
very well on the scale of the solar system or the galaxy or here on the scale of
everyday life on the surface of the earth, but it is a theory which when pushed
to very short distances and high energies begins to give mathematical
nonsense. If you ask questions like : What is the contribution to the force
between two particles due to the emission and reabsorption of quanta of
gravitational radiation of arbitrarily short wavelength?, the answer that the
theory gives you is that the force is infinite. I t's nonsense - it's clear the
theory is simply breaking down for very short wavelengths. How then do we
make sense of these two great revolutions of the twentieth century? How do
we combine our understanding of gravity with the quantum mechanics
which is supposed to govern nature in the very small?
In recent years there has been the development of a remarkable theory
called superstring theory, which perhaps now for the first time provides a
mathematically consistent theory of gravity. In superstring theory the
fundamental constituents of nature are seen to be not particles or waves, but
little strings, either opened or closed, continually joining and breaking
apart. Each string can vibrate in many modes and the vibrations of these
strings are supposed to stand in one-to-one correspondence with the various
species of elementary particles. This is a theory that apparently is free of all
the mathematical inconsistencies of all previous theories of gravity and ,
perhaps even more intriguing , this is a theory that not only allows a
mathematically consistent theory of gravity, but can't help it : You can't in
this theory imagine that gravity didn't exist. So we have taken three big
steps. First , Newton described gravity successfully in terms of an inverse
square law , but he could always have imagined that it wasn't an inverse
square law - it was just the data that forced it to be an inverse square law.
Then Einstein in general relativity developed a theory of gravity which
included Newton's as a limiting case, and which explained why in that
limiting case it had to be an inverse square law. But Einstein's general theory
of relativity still left open the question : Well why is there any gravity at all?
Why does space curve? Why isn't space just perl'ectly flat and there is no
such thing as gravity? Who asked for that? In the superstring theory we have
for the first time a theory in which gravity cannot be left out.
The history of this theory is rather amusing . It was developed as an
attempt to explain strong nuclear forces, at a time before the advent of
quantum chromodynamics. The mathematical physicists who developed it
at the end of the 1960s were terribly embarrassed at the fact that the theory
predicted the existence of a particle which looked like the quantum of
16 S. Weinberg
gravitational radiation, the so-called graviton, and yet it was supposed to be

a theory of the strong nuclear forces, and so they couldn't understand what
this graviton was doing in their theory. Quantum chromodynamics came
along in the early 1970s, and put the string theorists out of business for a
while. Some of them continued to beaver away in obscurity, not getting
promoted. (I speak here n ot with bitterness, but with embarrassment ,
because I was not one of them. In fact I was one o f the people who helped to
ignore them.) Just in the last few years it's been widely realized that
superstring theories offer a good chance of a unified theory of all particles
and forces.
These theories have not been tested experimentally, and we are
desperately hoping that society will continue to fund the efforts of particle
physicists to pin down our rather speculative ideas. The only way they can
be pinned down is by building large accelerators which will allow us to test
our ideas, and refine our ideas. After all, Newton didn 't invent the theory of
gravity without knowing about the existence of the solar system.
So it goes. Newton's hope, 'I wish we could derive the rest of the
phenomena of nature . . . ', has not been fulfilled , but we are working on it,
very much in the Newtonian tradition : the formulation of increasingly
comprehensive quantitative laws. From this high viewpoint , all that has
happened since 1687 is a gloss on the Principia.
Acknowledgement
Research supported by Robert A. Welch Foundation and NSF grant PHY
8605978 . I wish to thank Gerald Holton and Victor Szebehely for their
valuable help in the preparation of the published version of this talk.
3
Newton , quantum theory and reality
R O G E R P EN R O S E
3. 1 Newton's corpuscular-undulatory view of light

The supreme stature of Newton as a scientist cannot be doubted. As an
experimental physicist, he had superb natural skill, and was profoundly
ingenious, as well as being exceptionally careful in the construction and
execution of his experiments. As a mathematician he possessed
extraordinary power, and, indeed , had few mathematical peers over the
whole of history. In fact, it was very necessary for him, in developing his
scientific theories, to explore and innovate in mathematical ways, and he
uncovered many deep mathematical truths - the fundamental theorem of
calculus being but one. The Principia itself, whose three-hundredth
anniversary I am most honoured to be able to celebrate with this essay,
contained among its many propositions, multitudes of mathematical gems
and numerous profound mathematical ideas. In addition to these scientific
qualities, and most importantly for the development of his science, Newton
also possessed superlative instincts into the workings of Nature - instincts
which often fell hardly short of miraculous.
In one particular instance, concerning Newton's preference for a
corpuscular theory of light to a wave theory , the extent of his insight is most
intriguing - and perhaps somewhat debatable. The experimental evidence in
Newton's day was inconclusive, yet it would seem that the balance of that
evidence should have been in support of a wave theory of light rather than a
corpuscular one. Ironically, it was Newton's own discovery and painstaking
analysis of an optical interference phenomenon - Newton's rings - which
should have provided the strongest evidence in favour of the opposing wave
theory! (Other types of interference had been observed earlier by Grimaldi.)
Yet Newton stuck doggedly to a corpuscular view, and was led to a picture
of light quanta possessing both particle-like and wave-like properties. Can
18 R. Penrose
this apparent anticipation of modern quantum theory - a theory which was

not formulated until over two hundred years later - be other than the merest
fluke, resulting perhaps from Newton's obstinate refusal to give in to his
adversaries and accept their wave picture of light?
In his preface to the 1 952 Dover edition to Newton's Opticks, I . Bernard
Cohen rightly warns us not to make too much of Newton's apparent
anticipation of our present quantum-mechanical concepts. Yet, one cannot
but be struck by the fact that this master among scientists found himself
driven to a view of the physical nature of light that was not only seemingly
almost absurd, with its curious intermixture of corpuscular and undulatory
properties, but absurd in ways which seem close to the absurdities of modem
quantum theory! Is not this an indication that Newton possessed insights of
such power that they even gave him hints of revolutionary developments not
to occur until over two centuries later? On the other hand, perhaps the
similarity with quantum mechanics is indeed illusory. Is there really much
detailed correspondence between Newton's undulatory 'fits of easy
transmission and reflection' when his corpuscles of light interact with the
material of a refractive medium , and the wave-like aspects of photons in
their interactions with matter, as we understand them today? I would accept
that these analogies should not be pressed too far, and that the
correspondence between Newton's picture of light and that presented by
modem quantum mechanics would not stand up to much detailed analysis. I
can believe, also, that Newton possessed the mathematical wherewithal , so
that in principle he could have pressed forward with an entirely wave picture
of light, had he wished to do so , which would have been well in accord with
the observations of his day. Yet I think that it is likely that Newton did
indeed have powerful reasons for rejecting a purely wave picture of light;
and there was evidence available even at that time, vaguely discernible to a
scientist of Newton's sensitivity, for a view of reality in which undulatory
and corpuscular ingredients of reality co-exist at a fundamental level .
However, such reasons would have to have been ones that Newton could
not easily have articulated - or perhaps would not have wished to do so,
owing to his avowed dislike of purely 'hypothetical' contentions.
3.2 Stated reasons for rejection of wave theory

Let us first briefly examine Newton's stated reasons for rejecting the wave
theory. The main one seems to have been the rectilinear propagation of
light, and the fact that light does not bend into the shadow of an obscu ring
body. As we now know, from the phenomenon of destructive interference
Newton, quantum theory and reality 19
discovered by Thomas Young a century later, an effective rectilinear

propagation does indeed occu r for wave propagation when the wavelength
is sho rt, such as that which occurs with visible light - and light does indeed
bend slightly into the shadow. We do not blame Newton and his
contemporaries for not knowing this. Yet I suspect that if Newton had
preferred the wave theory for other reasons, it might not have been beyond
him to suggest something of the nature of destructive interference. (The
wavelengths of visible light were, in effect, known to him from his 'Newton's
rings' observations.) However, his preferences were for a corpuscular
theory , which rectilinear propagation seemed to support, so he did not find
himself compelled to examine this matter more deeply.
Newton's other stated reason for rejecting the wave theory of light was
that light has 'sides' - which was Newton's tem1 for linear polarization . The
observations of Erasmus Bartholin and Christiaan Huygens concerning the
phenomenon of double refraction in calcite, or Iceland spar, had led Newton
to conclude that a beam of light consists of components which are not
rotationally symmetric (i .e., they are polarized). This is, indeed, quite
incompatible with the type of longitudinal compression wave that Huygens
had suggested for light. However, a transverse oscillation, as was later
suggested by Young, would fit the bill. Apparently it occurred to neither
Huygens no r Newton that light might be a transverse oscillation. One can
again speculate that if Newton had had other strong reasons for preferring a
wave theory of light, he might perhaps have been driven to think of the idea
of transverse oscillations as the explanation of 'sides'. However, he had not
and he was not.
These two arguments for rejecting the wave theory of light, though
undoubtedly impressive at the time, must be counted as 'wrong' ; so if this
was actually all that Newton had to go on, we should , indeed , have to
consider that it was a 'fluke' that he found himself driven to a combined
corpuscular-undulatory theory oflight. However, it seems to me to be likely
that Newton may well have had various other reasons for disliking an
entirely undulatory theory oflight. One can only surmise what some of these
might have been . One possibility that has often been suggested is that
Newton favoured an atomistic view of Nature. Perhaps there is some force
to such an argument. If we are to believe that light is a wave motion, then
surely it must be a wave in some medium. On the atomistic view, this
medium must itself be composed of particles. This gives light a character
secondary to that of the particles of the medium. Yet, there is perhaps
something primitive about light that argues against it having such a
20 R. Penrose
secondary nature. If new particles are to be invoked in order to compose

such a hypothetical all-pervasive medium, is it not much more economical
to use such new particles as the constituents merely of the light itself?
Newton seems to have had some difficulties in coming to terms with such
an all-pervasive medium, and he discussed , at some length, the strange
hypothetical properties that such a medium would need to have (in Queries
18-22 of Book Three of his Opticks). Even when contemplating such a
medium, he preferred to have light constituted separately as particles of its
own . (Influences in the medium were imagined somehow to be able to
overtake the light corpuscles.) He also had some idea that an explanation of
motion under gravity might be found in terms of 'refraction ' by such an all
pervasive medium, and he presented a picture which even has something in
common with the ideas of general relativity ! (Perhaps, instead , Newton's
ideas were foreshadowing De Broglie's realization that matter itself
possessed wave-like qualities. Take your pick !) Newton appeared to have no
qualms about suggesting such a view of the natu re of gravity, while at the
same time holding to his picture of an inverse square law of force holding
between all pairs of particles in the universe. No doubt he regarded all these
pictures as highly provisional . He admitted to not knowing what the nature
of such a medil,1m might be (Query 2 1). The medium would need to have
some strange and powerful properties; and I suspect that Newton's view was
that the medium, if it existed, could not be material in any ordinary sense,
but would have to be of some mysterious substance quite different from that
of matter . Perhaps this was the 'absolute space' that Newton was really
groping for. Perhaps, indeed , if the modern ideas could have been explained
to him, Newton might not have been averse to viewing this medium in terms
of Einstein's picture of a 'flexible' spacetime continuum!
None of this implies that light must be corpuscular. Ordinary atomism
might require merely that the medium, if it exists, be corpuscular. It is just
economy, or perhaps aesthetic requirements, which , together with atomism,
would imply that light must be corpuscular. I wish to suggest that there may,
instead , have been another quite different motivation underlying Newton's
preference for a corpuscular view of light. This is the motivation from
Galilean relativity.
3.3 Newton and relativity

The reader may well object that there is not much evidence that Newton was
at heart a dynamical 'relativist' . He was certainly well aware that his
dynamical laws are invariant under change to a uniformly moving frame (as
well as under shift of origin and under rotation). This much he made clear in
his Principia. But what evidence is there that Newton at any time shared
Einstein's conviction that physics must be invariant under change to uniform
motion? The common view seems to be that, on the contrary, 'Newton
believed in absolute space' - with its decidedly non-relativistic connotations.
If space (as opposed to spacetime) is indeed absolute, so that it is meaningful
to talk about the same point in space at two different times, then there is an
absolutely preferred state of rest. It then becomes a 'fluke' that Newton's
dynamical laws are Galilei-invariant. There is no need to believe that this
fluke should continue to hold when additional ingredients of physics are
added to the basic dynamical laws. One such additional ingredient is the
velocity of the medium in which light travels, if it indeed be the case that light
merely consists of wave motion in some all-pervading medium. If we insist
that light is merely a wave motion, then we must give up Galilean relativity.
Of course , Lorentz, Einstein and Poincare have now taught us that there
is a way out of this. Light can still consist of a wave motion - this time, the
waves of oscillating electric and magnetic fields of the Maxwell theory - and
invariance under change to a uniformly moving frame can be maintained.
However, this requires a drastic revision of dynamical laws (for speeds
approaching that of light) and also an overturning of one's very notions of
space and time. It is thus, as we now know, that Galilean relativity can be
given up and replaced by another relativity - the relativity of Einstein and
Poincare.
I am certainly not attempting to suggest that Newton could have had the
absurd foresight to sense the need , or even the possibility of such a revision
of basic principles. However, if we do accept that this route was not
effectively open to him, then we see that Newton was presented with a clear
choice : give up Galilean relativity or stick to a corpuscular theory of light.
Now what evidence is there that Newton believed 'in his bones' that some
kind of dynamical relativity must hold? I contend that, despite what appears
to be a common belief to the contrary, such evidence does exist. I refer the
reader to Richard Westfall's fine biography of Newton (Never at Rest, 1 980).
On p. 147 Westfall describes how, in January 1665, recorded in private notes
to himself in a work referred to as his Waste Book, the youthful Newton
derived the laws of impact for two perfectly elastic bodies. (Huygen's slightly
earlier treatment of this problem had not yet been published .) Newton's
procedure was to consider the two bodies as a single system, and he came to
the conclusion that their common centre of gravity moves inertially whether
or not the bodies impinge upon one another. Realizing that every impact
22 R. Penrose
can be viewed from the special frame of reference of the common centre of
gravity of the two bodies, Newton appears to have used the impact
behaviour in that frame - in which the behaviour may be regarded as
'obvious' - and then assumed Galilean relativity to derive the elastic impact
law in the general case. This was more than twenty years before the
publication of Principia, but Westfall remarks, concerning the consequent
uniform motion of the centre of gravity, that 'Newton never forgot this
conclusion. It contained the first adumbration of his third law of motion , . . .'
- and it seems to have formed one of the central guiding ideas behind
Newton's dynamics, despite the fact that in Principia Newton chose to
derive his dynamics from different axiomatic principles.
In his account, Westfall does not draw particular attention to the fact that
Newton was implicitly relying on Galilean invariance to derive his impact
law,t although Newton's procedure cannot be used without such an
assumption. Without it, one has no right to 'pass to the special frame of
reference of the common centre of gravity' and to assume that the same
dynamical laws must hold in that frame as in the original frame. It seems to
me to be highly likely that by the age of 23, when Newton produced this
argument, some stroqg intuition concerning the relativity of motion was
already an essential ingredient of his thinking. It is, after all , a necessary
ingredient of the Copernican-Galilean mode of thought that the dynamics
that we employ at the surface of the Earth should be insensitive to the
Earth's uniform motion about the Sun . Newton was well aware of this, of
course. (It appears that, at least by 1666, Newton had read Galileo's
Dialogues, which appeared in translation in England in 1665.) It would have
t Towards the end of his life, Galileo had claimed a solution of the problem of the
dynamics of impulses. However, no account of this appears to have su rvived {cf. Galilei,
163 8 ; Dover edn , p. 293). It is intriguing to speculate whether Galileo might have been
guided along similar lines of reasoning.
Although Huygens's treatment of this problem seems to have been a little different
from that of Newton, he also used Galilean invariance {cf. Mach, 1972) . It is striking to
note that Huygens did not believe that this Galilean invariance extended to light, in the
sense that I mean it here, since he strongly promoted the wave theory ! Huygens's view
seems to have been that the individual particles of the medium might be subject to a
Galilei-invariant dynamics even though the medium as a whole might define a rest
frame. However, this viewpoint attaches an 'accidental' rather than a 'fundamental' role
to this medium - and therefore to light itself. It also presupposes that the Galilei
invariance of the local dynamics of macroscopic bodies is insensitive to the presence of
the medium - another 'accidental' {and, moreover, highly improbable) feature. While the
example of Huygens shows that it was possible for a great scientist of the time to hold to
the wave theory and also to be a professed relativist, this is not totally being a relativist
in the 'deep' sense that I mean here.
been natural for him to suppose that other physical properties (such as the
behaviour oflight) should also be governed by the same beautiful invariance
principles. Perhaps, at that time, he had not explicitly formulated , even to
himself, what such 'invariance principles' might entail , but my guess would
be that he knew pretty well within himself what he was about. His treatment
of the problem of impulses seems to me clearly to indicate that he
instinctively felt that a relativity principl e must be intrinsic to Nature's ways.
At roughly the same age as he studied the impulse problem , he was also
becoming profoundly interested in the nature of light. On Newton's own
accou nt, he 'had the Theory of Colours' in January 1666 (Westfall , 1980,
p. 1 56). He may even have had some inkling (as had Galileo before him - cf.
Drake, 1957, p. 278 , concerning a remark made in The Assayer) that light
has some profound role to play in governing the minute structure of matter.
One can readily imagine that Newton's great intuition guided him to believe
strongly in some kind of relativity principle encompassing both the
dynamics of matter and the behaviour of light. This intuition would have
gu ided him surely to a corpuscular rather than to a wave picture of light.
That would have been in the years when his profound views on the nature of
light were first being formed. These years must have set the stage for all his
later thinking. When at a later time in his life he became fully confronted by
the experimental facts of optical interference phenomena, he would have
been reluctant to give up the intuitions that he had built up during this
formative time. I suggest that in his 'bones' he knew that light must be
co rpu scular, even though in later years the good reasons for these early
intuitions may well have receded from him. And unlike the stated reasons
that Newton later put forward in his Opticks in favour of a corpuscular
picture of light - which, technically, were 'incorrect' - this unstated one, that
a relativity principle mu st hold, was, I am claiming , profoundly 'correct' !
When Newton came t o write his Principia, some twenty years after these
intuitions must have been formed , he needed to place his logical
development on a clear footing . To state his basic laws in a definite
mathematical way, he needed the framework of 'absolute space' so as to be
able to formulate things precisely. Even with the abstract mathematical
machinery of today, it is n ot so easy to give a totally Galilean-relativistic
treatment of space, time and dynamics right from the start. It can be done,
but one needs an absolute spacetime, rather than an absolute space. (See
Arnol'd, 1 978, for an excellent modern such ' treatment .) Newton did take
pains to be completely clear that Galilean invariance was indeed a featu re of
his laws (cf. Scholium 5, following Definition VIII), but this invariance was
24 R. Penrose
not a necessary principle. It might even be that the technicalities involved in

writing his Principia guided Newton away from what I believe must have
been his early relativistic intuitions. Once Galilean relativity has become a
consequence of the basic laws, rather than a basic principle, there appears to
be nothing against adding to those laws something - such as an all-pervasive
medium - which does not share in that Galilean invariance. Accepting all
the profound influence, power and comprehensiveness of Principia, and
accepting all its magic in deriving from such simple axioms so much of the
dynamical workings of the world , one can still contemplate the case that its
very strength of mathematical structure could have obscured a youthful
insight - an insight that had to wait nearly two and one half centuries to be
profoundly reinstated in a new form by Einstein in 1 905 : that one should
take relativity as a principle, rather than as a seemingly accidental
consequence of other laws. (See note added in proof, p. 49.)
3.4 Newton's route to an undulatory-corpuscular picture?

I am conjecturing that , despite Newton's speculations in his Opticks,
concerning the possibility of an all-pervasive medium , and despite the
seeming implications, in Principia, that he might be happy with an absolute
space, Newton carried forward from his youth a deep insight into the
relativistic nature of physics. Indeed , he well recognized that his space and
his medium, if they 'existed' in any ordinary sense, would have to have
profoundly peculiar properties. However, it would be far too much to expect
of him that the realities eventually revealed by Maxwell concerning
electromagnetism and light, and particularly those revealed by M inkowski
and Einstein concerning the essentially four-dimensional structure of
spacetime, could have been discernible by him. That would have been a
quite unbelievable route for him to follow . Instead, he chose the alternative
Galilean-relativistic, but , for the time, equally valid , route. His early
relativistic insights, I am suggesting, may well have been largely responsible
for an intuitive belief in a corpuscular picture of light. Once he was firmly
down that road , and having to come to terms with optical interference
phenomena, he was led almost inevitably (and , in a sense, 'correctly') to a
picture of light in which corpuscular and undulatory features had somehow
to coexist, leading to an eerie foreshadowing of quantum wave-particle
duality over two hundred years before its time.
It may well be that there were other 'valid' hints that a combined
undulatory and corpuscular picture had to be entertained . Newton's
interest in the colours of actual objects, and his alchemical work concerning
the transmutation of substances, must have convinced him of a deep

interrelation between light and matter. He certainly speculated that matter
and light might be convertible into one another (Opticks, Query 30). One is
reminded of Einstein's early struggles concerning the absorption and
emission of light (see Pais, 1980). It was not only Planck's explanation of the
black-body spectrum which disturbed Einstein. The whole idea that
individual atoms could discretely pick up energy from what, at that scale of
things, would have been a barely oscillating and nebulous medium seemed
almost preposterous. Einstein's intuitions and insights guided him to the
view that , despite the fact that light is waves, light must also be particles :
indeed Natu re is preposterous! It is not inconceivable to me that a few of the
many worries that so haunted Einstein in the early twentieth century might
already have been vaguely discernible to Newton over two hundred years
earlier!
3.5 Quantum mechanics

For the rest of this essay, I shall depart from such speculative considerations
of pseudo-history and concentrate on this very preposterous nature of
reality that the observational facts of modem quantum physics have
presented us with. The questions of quantum reality are closely related to
some of t hose we have just been discussing. We shall find that these
questions will lead us into some speculative considerations of a different
kind !
Let us remind ourselves of one of the very basic rules of quantum theory,
namely linearity. If I t/I > and I </>) are two vectors representing quantum states,
then any quantity of the form
A ll/I> + µI </> ) ,
where A and µ are complex numbers not both zero, also represents an
allowable quantum state. Recall that multiplying a quantum state-vector by
a non-zero complex number yields a state-vector representing physically the
same quantum state as before, while all other state-vectors describe
physically distinct quantum states. Thus, taking all the different possible
ratios A/µ (including w ) , we are provided with an array of physically
allowable alternative states, obtainable by quantum linearity from the
original two states. The abstract space of such complex ratios has the
topology of a spherical surface - the Riemann sphere.
This sphere is implicitly present in any two-state quantum system.
Sometimes it has a very clear geometrical role to play ; sometimes this role is
26 R. Penrose
rather subtle. It is clearest for the case of the spin-state space of a spin one
half massive particle, say an electron. Here each physically distinct state of
spin corresponds to a definite direction in space. This direction is the one
with the property that if the spin of the electron is measured in that direction,
then the answer to the measurement is : yes, with certainty. In the case of
polarization states of a photon , the geometrical role of this sphere is slightly
less direct. Suppose that the photon is moving upwards. Then the north and
south poles correspond, respectively, to the states of right-handed and left
handed circular polarization. The points around the equator correspond to
the states of linear polarization (Newton's 'sides'), but since a linear
polarization state corresponds to an undirected line element , the two
opposite directions of this line element must correspond to the same point
on the equator. More directly, it is the 'square root' Riemann sphere for
()./µ)1 1 2 which has a clear-cut geometrical correspondence with polarization
directions (Stokes vectors).
Sometimes the geometry is not at all direct , such as with the various linear
superposition states of two different positions of a particle. Suppose a
particle is passing through a pair of slits A and B . If the particle is at A , then
we represent this by the north pole of our Riemann sphere. If it is at B, then by
the south pole. The various other points of the sphere correspond to states in
which the particle is partly at A and partly at B. As we know , it is not a
question of mere probabilities of the particle being at A or B, but of
amplitudes of it being at A or B whatever that might mean ! The space of
-
different possible ways that the particle can be located as it passes through
the pair of slits has the structure of a sphere - not a line -segment , as would be
the case with probabilities (the range of probabilities being the real-line
segment (0, 1]).
3.6 Physical reality

We are here beginning to be confronted by the odd picture of reality that
quantum mechanics presents us with. Do we really believe that the particle
can be in two places at once, with complex weighting factors? Our intuitions
are not so stretched in the case of the spin states of the electron . In that case
we can visualize the electron as 'spinning' about an axis which has some
definite direction in space. The photon, too , can be thought of as having
some definite geometrical polarization state. Some people feel uneasy with
such clear-cut geometrical pictures of quantum states. They say that we
should not try to form a picture of reality when applying the rules of
quantum mechanics : just follow the rules of the quantum mechanical
formalism, do not try to form pictures and do not ask questions about
reality ! This seems to me to be wholly unreasonable. Physics , after all ,
constitutes our best way of groping for the true nature of the real world in
which we find ourselves . Some would have us believe that physics is to be
regarded as merely providing a means of effective prediction about the
future. It is true that one of the most powerful shifts in attitude in science
came about with the methodology of Galileo and Newton : form a
mathematical theory from which predictions can be made and test these
predictions against observation. Do not ask what it is that constitutes the
substance of things , nor why things behave as they do . Just ask how they
behave and try to form an elegant mathematical structure that mimics that
behaviour as closely as possible. ('Hypotheses non jingo '!)
However, the power of such methodology should not obscure the fact that
we are still , nevertheless, groping for the realities of the physical world. The
methodology is powerful because it enables us to clear from our minds a
good many preconceived and incorrect notions concerning the nature of this
reality . It does not, and it should not, lead us to believe that we can dispense
with the notion of physical reality altogether - as some would seem to have
us believe! What, after all , is the point of making 'predictions about the
future' unless we are allowed to assume that there is some reality about this
future whose state we are proposing to predict? Despite the operationalistic
nature of the methodology of Galileo and Newton, their dynamical theory
provided a picture of 'reality' which was very much clearer than that which
had gone before. In fact , this picture has become so 'clear' to us that there is
now a great reluctance, even after we have been forced to accept the physical
existence of quantum phenomena, to believe that 'reality' can take any other
form .
It is clear from Newton's Queries in his Opticks, that he was , indeed , well
prepared to speculate on the nature of physical reality and that, in the case of
gravity, the existence of a wonderful mathematical theory did not stop him
from speculating on 'causes' (cf. his refractive all-pervasive medium referred
to above) . 'Hypotheses non jingo ' did not apply to Opticks (nor, indeed, did
it apply to Newton's earlier thinking) ! It seems likely, also, that Newton's
picture of reality was actually not very close to the 'clear' picture of reality
that we have subsequently built up from a familiarity with Newtonian
dynamics.
3.7 Reality of the state vector

Let us return to quantum mechanics and to the strange pictures that it
28 R. Penrose
presents us with. We have seen that, with electron spin, our geometrical
picture of the quantum state-vector is not at all an unreasonable one.
However, even here, many would say that we are deluding ourselves, and to
form such a picture is misleading . Somehow the electron has only two ways
in which its spin can point, not a whole spherical continuum of ways. Any
experiment that we may perform on the electron to measure its spin can give
us but one bit of information : the spin can turn out to be one way, or it can
turn out to be the other, but it can be nothing in between . That is the way
quantum mechanics works.
However, when there are just two such 'alternatives' , the space of st�tes is
indeed a whole Riemann sphere. We cannot say which the alternatives are.
They might be up/down or north/south or east/west or anything in between .
There is nothing to choose between any of these pairs of alternatives. But
given any particular state of spin of the electron , there will be precisely one
direction in space for which the state gives certainty that the spin is in that
direction. Given the state, there is a 'reality' about the direction in which the
state points. An observation might be performed measuring the spin in that
direction and , if so , the state has to be prepared to say 'yes', with certainty , to
that measurement. Somehow the state has to 'know' that direction, even
though there is no experiment which can be performed to determine which
direction in space it actually is. So long as we are sticking to standard
quantum mechanics, the different possible 'realities' for the states of spin of
an electron are indeed the points of the Riemann sphere. This is for a system
with just two alternative states. In a sense, the whole Riemann sphere
'counts' as just two !
For massive particles of higher spin there is also a corresponding
geometrical picture for the states. Take the spin to be n/2 , then in place of the
Riemann sphere we have a complex n-dimensional projective space of
different possible physical states. But we can still use the Riemann sphere to
describe the individual states. Each physical state of spin n/2 co rresponds
uniquely to an unordered set of n points on the sphere of directions.t A
characterization of these points is that if the spin is measured in that
direction then there is zero probability that the spin turns out to be totally in
the opposite direction . Again , the state is uniquely characterized by the fact
t I am told that this result is due originally to Majorana, but I do not know of a
reference. The result is not hard to see using 2-spinors (cf. Penrose and Rindler, 1984 ,
p. 162). If ijJ AB N is an n-index symmetric spinor representing the state, then it has a
canonical decomposition ijJ AB . N = r:x. (A f3B . . vN ) (round brackets denoting
. . .
vN .
. .
symmetrization) , and the n directions are those represented by the spinors r:x. A , /3B ,
.
• • . ,
that if measured in any one of these n directions, then with certainty it

cannot spin totally the other way. The state has to 'know' that it must
respond this way in the event that any one of these measurements is carried
out , so, up to proportionality , it is again objectively characterized.
The same objectivity holds, in principle, for any quantum state whatever
so long as the rules of quantum mechanics are presumed to hold rigorously.
Suppose the state-vector is I t/I ) . Then we can consider making an
observation corresponding to the observable
Q = ll/J)(l/11 .
The state I t/I> is the only one, up to proportionality, for which the observable
Q yields the result 1 with certainty. The state must 'know' that it has to
produce this result in the event that the observation Q is actually performed .
This is a completely objective property, so the fact that the state is
(proportional to) I t/I ) is an objective property of the world.
I am expressing a point of view here which, for some reason , is not often
maintained . One frequently hears the ?pposing view that 'the state-vector
merely expresses our state of knowledge about a system', or 'the state-vector
is expressing a property of ensembles of systems rather than of a single
system' . However, the point of view that I am expressing is that the state
vector is clearly defining an objective property of a single system. It is not a
'testable' property of that system in the sense that we can perform an
experiment on it to determine what its state I t/I ) actually is, but it is an
'objective' property of that individual system in the sense that the state of
that one system is characterized by the results of experiments that one might
perform on it.
3.8 Quantum non-locality

Why is this objectivity not something that is more frequently expressed by
theorists? The answer to this question is not really clear to me, but I suspect
the reason has to do with the strange picture of objective reality that it
presents us with. Take the situation considered above in which we had a
particle passing through two slits A and B. The objective reality of the
system , I claim, actually consists of the particle being partly at A and partly
at B, both together in some complex linear combination (since for each
choice of complex linear combination of being at A and being at B there is an
experiment that could be performed which gives the answer 'yes' with
certainty only for that particular combination). However, we like to think of
particles as tiny point-like objects, not as things that can be in two places at
30 R. Penrose
once. Moreover, we may choose to measure whether the particle is at A , or

to measu re whether it is at B, and if we find it at one place we cannot find it at
the other. Somehow, if we perform such a measu rement, we find that at the
next moment the particle is either at A or B, so we like to think that, a
moment before, the particle really was in one place or the other, but because
of some kind of funny statistics which holds at the quantum level the
alternatives must be weighted with complex rather than real coefficients. If
we take the realistic view of the state-vector that I maintain, in view of the
above discussion, is really almost forced upon us, then we must accept a
degree of non-locality : finding the particle at A instantly forces it to be not
found at B and, conversely, not finding it at A instantly forces it to be at B.
This latter type of observation is an instance of the particularly confusing
situation referred to as a 'null measurement'. The mere fact that a detector
somewhere has not registered the presence of some particle forces the state of
that particle to be different from what it was before in some region perhaps
distant from the detector itself. It is especially puzzling that the absence of a
physical interaction can affect the state of a particle in some distant region !
This non-locality is perhaps even more striking when there is more than
one particle involved . The state-vector may consist of linear superpositions
.
of different arrangem ents of the particles. Observing the state of one of the
particles may instantly force the others into some state correlated with it,
even though they may be well separated from it. As an instance of this sort of
thing, let us consider a variant of Bohm's version of the Einstein-Podolsky
Rosen thought experiment . Suppose that we have a particle in a spin-zero
state which decays (non-relativistically) into two particles each of spin one
half. Suppose also that there is no orbital contribution to the angular
momentum. We can express the various spins in terms of eigenstates of spin
in the vertical direction . Then we can think of the initial spin state I O) to be
decomposed , after decay, into the difference between two states, one state
having one of the particles spinning up and the other down , and the other
state having the spins of the two particles both reversed :
I O> = Ii> l l> - l l> I i>·
However, there is nothing special about the choice· o f the vertical direction
here. We could equally well have chosen left/right :
IO > = I +-- ) I -+ ) - I -+ ) I � ) ,
or, of course, any other pair of opposite directions. Neither of the resulting
particles has, by itself, a state. Only the pair considered as a whole has. If we
observe the spin of one of these particles to have some particular direction,
then the other instantly acquires an 'objective spin state' , namely the state
corresponding to spin with certainty in the opposite direction . This applies
equally to any di rection that we may choose. Moreover, there is no
requi rement that the two particles be at all close to one another. In principle
they could be hundreds of miles apart . By observing the spin of one of the
particles , we instantly put the other particle into a spin state whose direction
is fixed by our direction of observation on the first particle! (We note that no
message can be sent from one particle to the other by this means .) This is a
decidedly non-local picture of reality.
I am regarding the total state on the right-hand side of these equations as
being 'real '. But how else are we to regard it? We are always at liberty to
choose to measure the spins of neither particle but , instead, to reflect both of
them - carefully, so as not to disturb their spins - back to thei r original
position so that they can recombine as a spin-zero state. The spin of the
recombined state must indeed be zero with certainty, and the possibility of
observation of this final spin assigns a 'reality' to the non-local spin state of
the combined system. Other situations of this same general nature exhibit a
similar non-locality. This is particularly striking when the two emitted
particles are photons since the effect of observing one of them would, if it
were a signal in the ordinary sense, have to travel faster than light in order to
reach the other one. The experiments of Freedman and Clauser ( 1972) and
Aspect ( 1 976) and coworkers, etc . , have shown that a non-locality of this
type (over a distance of some twelve metres, in the latter case) is an actual
featu re of the world we live in and not just a theoretical fiction.
3.9 Quantum mechanics and macroscopic physics

We see that this quantum view of reality is very different from the one that
we have become accustomed to from classical physics , where particles can
be only in one place at a time, where the physics is local (except for action at
a distance) and where each particle is a separate individual object which,
when it is in free flight , can be considered in isolation from any other
particle. All these classical conceptions must be overturned once we accept
the reality of the state-vector. It is perhaps little wonder that most people are
reluctant to do this. However, the quantum view of reality is more
worrisome even than this. For the linear superpositions between different
particle positions can get coupled in with macroscopically differing states of
macroscopic objects. In that case, according to the strict rules of quantum
mechanics (and assuming that no 'observation' is deemed to have been
made) we must carry o ver the complex linear superpositions to the
32 R. Penrose
macroscopic objects themselves. Thus we can be presented with a situation

like that of Schrodinger's cat, where we seem to have to consider states
representing linear combinations of a dead cat and a live cat, with complex
coefficients. There is a very reasonable reluctance to accept such a state as
actually describing reality ! Surely, at that level the cat is either dead or alive.
However, there still remains the possibility, in principle, that one might
perform the observation described by Q = l t/1) ( 1/1 1 , as suggested above, that
requires objectivity for the state II/I> whatever it is. In this instance, where II/I>
represents some complex linear combination of a dead cat and a live one, it
might well be objected that the observation described by Q is quite out of the
question practically. But quantum theory, as it stands, has nothing to say
about what is 'practical' or 'impractical'. To place some restriction on which
operators Q are allowed as observables, and which are not, would be to
introduce a change into the theory.
Some people take the view that a quantum linear superposition at that
level is still perfectly in order. A conscious human observer could also be in
such a superposition but, because of some quirk of the nature of conscious
perception, his consciousness would 'split', so that each state of awareness
would perceive just one of the macroscopically distinct states under
superposition , and would not be aware of the actual superposition. This (or
something like it) is the many-worlds view of Everett and his followers.
Personally, I should be most reluctant to go to such lengths to save the
superposition principle for macroscopic bodies. Not least , it is worrisome to
have to rely on a yet undiscovered theory of consciousness in order to find
agreement between the picture of things presented by theory and what is
observed in practice. It seems to me that something must go wrong with the
superposition principle a good deal before this. Macroscopic physics is like
classical physics, where particles occupy single localized places and behave
as individual separate entities.
3.10 Linearity and time-evolution

So far, I have not discussed the time-evolution of quantum states . As we
know, there appear to be two completely different modes of evolution in
quantum theory - and this decidedly odd fact has, no doubt , added to the
reluctance that many people may feel about taking the state-vector as
actually representing reality. On the one hand, we have unitary evolution
(Schrodinger or Heisenberg, according to one's description). This is
completely deterministic, as in classical physics, and it preserves quantum
linearity. On the other hand , we have 'state-vecto r reduction', which is
probabilistic and does not preserve quantum linearity. If, as I am proposing

here, one takes a realistic view of the state-vector (and supposing that one
rejects the many-worlds view - or even if one does not), then one must come
to terms with these two different modes of evolution.
Some would take the line that state-vector reduction is an illusion which
appears to arise because one is ignoring the complicated interconnections
between the system under consideration and the outside environment. The
claim is, on this view, that unitary evolution would hold if we could take
everything into account, but, since we are ignoring the outside, we are led to
an approximation which appears like a reduction process. I have to say that
I do not find this viewpoint at all plausible. I really do not see how the linear
superposition of Schrodinger's dead and alive cat can get resolved to one
alternative or the other, simply by taking the environment into account. The
best that one can hope to do with this type of procedure is to show that
somehow the cat can be treated in practice as some probability-weighted
mixture of alternatives. It cannot explain how the reality of the situation can
shift from a complex linear superposition to alternatives. Unless we change
the theoretical structure of quantum mechanics at some stage we still have
the possibility in principle of performing the measurement Q referred to
above, so 'reality' must remain the complex linear superposition !
Another su ggestion is what is known as the 'Wigner view' (Wigner, 196 1),
according to which unitary evolution holds until consciousness gets
involved in the physical system. At that stage, some non-unitary evolution
takes over and state-vector reduction objectively takes place. Again I am
worried that a theory of consciousness seems to be required in order that the
theory can be used . But a stronger objection, to my mind , is that the view
seems to assign a privileged role to those corners of the universe where
consciousness resides - and they may be very rare indeed . The objective
evolution of the state-vector would be quite different in those corners from
elsewhere. The effect might not be 'observable' but the asymmetry in the
pattern of time-evolution presents a disturbing view of reality.
My own picture is different from any of these, though it shares with the
Wigner view the idea that unitary evolution is to be supplanted, under
certain physical circumstances, by an objective evolution which closely
approximates state-vector reduction. However, I am not suggesting that
consciousness is needed for this. The reduction process would normally be
completed long before the results of the reduction are perceived by any
conscious observer. Indeed , an actual conscious observer would normally
be playing no role whatever in the reduction process. Instead , I am
34 R. Penrose
suggesting that, at a certain level of scale, a failure of quantum linear

superposition becomes important, and unitary evolution is taken over by
some specific non-unitary procedure. It might be that this procedure can be
described by some non-linear (and non-local) differential equation, non
linear instabilities arising at a certain scale, so that the system then 'flops'
into one or the other of two macroscopically distinguishable states. On the
other hand, the procedure might have to be described in some quite new
way. Non-linearity is essential, since the conventional reduction procedure
does not preserve quantum linearity ; but it is not clear to me that a
differential equation is the right thing to look for. Also I do not wish to
prejudice the issue as to whether the reduction procedure should be
genuinely stochastic or whether there should be some underlying
determinism which just has the appearance of randomness. The linearity of
quantum mechanics would , on this view , be merely an approximation, albeit
an excellent one at the scales where quantum mechanics has been tested .
Many people feel uncomfortable about dispensing with quantum
linearity, rightly pointing out that complex linearity is the source of so much
of the extraordinary mathematical elegance of conventional quantum
theory. To this I can only respond that we have a precedent. The remarkable
mathematical elegance of Newton's theory of universal gravitation, as so
magnificently expounded in Principia, also owes much to the fact that the
total gravitational force at any point is a linear function of the sources. Yet
when this theory was finally superseded by a more comprehensive one,
namely Einstein's general theory of relativity, it was by an astonishing non
linear theory of even greater mathematical elegance! I would anticipate that
the appropriate theory which replaces our present quantum ideas must,
when it comes, exceed even quantum mechanics in its mathematical beauty,
economy and grandeur. Unfortunately, I am in no position to make clear
suggestions as to what form such a theory might take. However, I am
making a definite proposal as to the kind of physics which should , in my
view , be relevant. This proposal leads us to a fairly concrete suggestion,
enabling us to give, in principle, very rough estimates as to when linear
evolution should fail and be taken over by some non-linear procedu re
resembling state-vector reduction.
3. 1 1 Quantum gravity and time-asymmetry

The new physical ingredient which must be added to quantum theory is, I
claim, general relativity. The new physical theory we are seeking is thus a
form of quantum gravity. As of now, no satisfactory theory of quantum
Newton , quantum theory and reality 35
gravity exists. Ideas abound and vast calculations continue to be performed ,

but there is still no adequate theory. My own ideas concerning the form that
this theory should take - and even concerning the very purpose of such a
theory - seem to differ from those of most workers in the field . Most people
seem not to be concerned with trying to alter the structure of quantum
mechanics. On the contrary, the impulsive reaction has generally been to
modify the structure of general relativity instead (to supergravity , more
complicated Lagrangians , higher dimensions, or strings), as soon as the
encounter between these two great theories runs into difficulties!
Why do I believe that quantum theory is likely to require revision when
unified with general relativity? Aspects of my views on this question have
appeared in other articles (Penrose, 1976, 198 3 , 1986) ; also, I refer the reader
to other related viewpoints which had been put forward earlier (Karolyhazy,
1966 ; Karolyhazy, Frenkel and Lukas, 1986; Komar, 1969). Part of the
reason for believing that the true quantum gravity theory may entail a
change in the structure of quantum mechanics is to be found in the very
difficulty that one encounters in trying to 'quantize' general relativity in a
standard way. Often these difficulties seem to confront one with
foundational problems concerning quantum observations. I do not propose
to go into a discussion of these here. Many of these arguments are familiar in
the literatu re (cf. Wheeler, 1964; Kuchar, 198 1). In my opinion there is a
considerable force behind these arguments. Attempts to build up an S
matrix theory of gravity from perturbation series starting from flat
spacetime, for example, simply do not face up to many of these long
standing foundational problems.
I wish to stress, on the other hand , a point of a different nature - one which
might even be called an observational fact of quantum gravity ! One of the
few situations in nature where it is generally accepted that quantum gravity
effects should be important is in governing the physics at spacetime
singularities (or, at whatever quantum-geometrical regions are to replace
such classical singularities). From the assumption that cosmic censorship
holds, we can deduce that these singularities can be divided into two classes :
past singularities (out of which particles can emerge) and future singularities
(into which particles can disappear). (See Penrose, 1978.) The one past
singularity that we know of is the big bang. Future singularities are expected
to arise inside black holes and perhaps in the big crunch (if the universe
eventually recollapses to a singularity). Now the big bang was very far from
being a 'generic' singularity. We know this both from the observed
uniformity of the early universe and from the very existence of a second law
36 R. Penrose
of thermodynamics (Penrose, 1979). (This argument is not affected by

current considerations of 'inflation'.) If we imagine the big bang time
reversed, we see more clearly how special it really is. In a generic big crunch,
we expect many black holes congealing finally into one very complicated
final singularity. The geometry close to the end would be extremely irregular
and not at all close to the highly uniform and ordered singularity that is
observed to be the case with the big bang. In the final big crunch (if it occurs
- or, in any case, as a theoretical consideration) we have an irregular,
extremely high entropy singularity. To a somewhat lesser (but still very
considerable) degree, the situation is similar for the singularities inside black
holes ; each such singularity would be of a very high entropy type. However,
the big bang was a singularity of a uniform and very low entropy type. The
lowness of this entropy lay in the fact that the gravitational degrees of
freedom were somehow initially not 'thermalized' with the matter and .
moreover, have not much become so because o f the weakness o f the
gravitational coupling. As time progressed, the matter in the universe has
gradually been able to take advantage of this imbalance and clump
gravitationally. Our Sun, itself, is a result of such gravitational clumping
and has become the hot spot in the sky that all life on Earth ultimately relies
upon as its source of low entropy.
Using the Bekenstein-Hawking formula for black-hole entropy , we can
estimate the entropy that a final big crunch would have, and hence deduce
how extraordinarily special the big bang actually was. The answer is
remarkable : for a closed universe with , say , 108 0 baryons, we find that the
big bang was special to the degree roughly
one part in 10 1 o m
where this figure refers to the region of phase-space volume that corresponds
to a big-bang singularity of the uniformity that is observed in the universe as
we know it. For a larger number of baryons, this figure would be even larger.
Allowing for inflation makes virtually no difference to the argument.
The specialness of the big bang can be accounted for if we assume that for
some reason - presumably as a consequence of the quantum gravity theory
that we are striving for - the Weyl curvature is necessarily zero (or at least
very small compared with the Ricci curvature) at any past singularity . The
same constraint cannot apply to future singularities, for, if it did, we should
not have a second law of thermodynamics. Moreover, there would then be
all sorts of teleological behaviour preventing the formation of black holes .
We see that this one 'observed' consequence of quantum gravity is grossly
N ew t o n , qua n t um th eory and rea li ty 37
t i m e-asym m etrical ! F rom t h i s w e can deduce t h e i m p o rtant m o ral : th e true

q ua n t um gravity must he a t ime-asymm etrical th eo ry .
3. 1 2 Tj m c-asymm ctry of sta te-vecto r reduction

No tice that this i s n o t at all the expected concl u si o n if we think of quantum
g ravity as a ri si n g as the result o f the applicati on of some stan dard q u antum
m echan ical p roced u re (wh ich would b e time-symmetrical ) to standard
general rel ati vity ( wh i c h is ti me-symmet rical ) . W h at i s n eeded is some n ew
p roced u re w h ich al l o ws fo r some time-asymmetry .
N o w , state- vecto r red ucti o n - o r the m o re compl ete p rocess that
u nderl i es i t - i s , J c l ai m , a time-asymmetri cal p rocess. In the fi rst place , the
rea l i stic view o f the state-vecto r that 1 h a ve been p ro m o ting i tself en tails a
t i m e-asymmet ry that is n o t n ecessarily p resen t in other attitudes to
q u antum m echani cs . F o r we n o t e that i mmediately aft er an obs ervati on i s
m ad e , t h e sta te- vecto r i s i n an eigen state of t h e rel evan t Hermitian operato r ,
but th i s d o es n o t gen eral l y apply to the state imm ediately before an
o b se rvati o n . (Ti me-sym m etry fo r q u a n t u m evol u ti o n can b e partial l y
rei n stated at the expense o f the real istic vie w , wh e re o n e regards transition
probab i l i ti es as p roviding the only observati o n al aspects of q u antum
theory ; sec , for exam pl e , Pen ro se, 1 979, p. 584. H owever, I am n o w
d i s p u ting m y earl i er acq u iescence in the c o m m o n conten tion that t ran sition
p ro babili ties are time- reversible - as we shall see in a m omen t ! ) N o te that
t h i s p i ctu re is tied u p with the fact that our real istic view seemed fo rced upon
u s because o f our p resu m ed freed o m , i n p ri nci ple, to carry o u t a q u antum
m cchan ical m easu remen t c o rrespo n ding to the Hermitian operator of our
ch o i ce , here the o pera t o r Q = l /t ) ( /t l
i i . The state has t o b e 'prepared fo r' the
p o ssibility that th i s m eas u rement m ight 'at any m oment' b e perfo rmed o n it.
T h e use of such an a rgumen t i n the p ast direction of time seems to carry
m u c h l ess weigh t , since o n e docs n ot seem to h ave the freedom to carry o u t
' u n expected ' ex peri m ents i n t h e past ! One thinks intuiti vely that t h e state
'al ready k n ew ' what was done to it in the past, b u t 'cannot know' what we
m ig h t u n expected l y do to it in the fu tu re . Of co u rse these feelings may b e
en t i rel y i n ap p ro p riate and i l l u s o ry . F o r very minute i n tervals o f time, I
would m ysel f n ot gi ve t h em m u ch credence . Yet , fo r a state which has
persi sted fo r a while, I can n o t help feel ing that there i s someth ing i n the
'co m m o n sense' attitude that the state is being go verned by what was done
to i t i n the past rath e r than b y what i s going to b e done to i t in the fu tu re .
(Th i s u se o f 'go ve rned ' h as n o t real ly m uch to do with t h e d eterminacy/
i n determ i n acy q uest i o n . 1 am referring, here , to the req u i rement that the
38 R. Penrose
state be an eigenstate of the past measurement immediately after it was

performed , but not an eigenstate of the future measu rement immediately
before if will be performed.) For the moment, at least, I am allowing this use
of 'common sense' to instruct us that ou r objective state-vector must indeed
behave in a time-asymmetrical way under the reduction process.
This time-asymmetry seems to be bound up in some way with our
'freedom of choice' to perform measurements. It is probably also , therefore,
bound up with the second law of thermodynamics - although the precise
relation between this question of 'choice' and the second law is, to my mind,
somewhat obscure. However, there is a second aspect of time-asymmetry in
the 'measurement process' which, in a certain way, seems also to be
connected with the second law. Imagine a two-state system , with an
orthogonal basis for states that I symbolically write as I i ) , l l > · Let us
assume that the system is initially prepared in the state I i ) . Suppose that a
measurement is then performed on the system , the eigenstates of the
measurement being I -+) and I +- ). (This need have nothing to do with spin. I
just mean these to be orthogonal states different from the original pair.)
There will be some resulting probability that the outcome is I -+ ) and some
probability that it is I +- ) . Two possible universe histories that must be
allowed by the · reduction procedure are therefore :
first I i> and then I -+ )
and
first Ii> and then I +-) .
Now if this procedure were time-symmetric , it would have also to allow both
first Ii> and then j -+)
and
first l l> and thenI -+ ) ,
if we assume that the final state i s given a s I -+ ) , and al so both
first Ii) and then I +-)
and
first l l ) and then I +- )
for a given final state I +- ) (where the words 'first' and 'then' are .being used
in their normal temporal sense), the same probabilities arising as before (in
reverse order), where now it is I -+ ) , or I +- ) , that is, 'given ' in the time
reversed reduction procedure.
Now, in practically any plausibl e experimental set-lip, the input I i ) , and
the two alternative outcomes I -+ ) and I +- ) , are all perfectly reasonable
Newton , quantum theory and reality 39
things to occur. However, l l > will normally be 'absurd' - at least in the sense
that it represents an extreme improbability which violates the second law -
this probability being nothing like what would be predicted by the standard
quantum-mechanical computation . For example, one might have a Stern
Gerlach apparatus preparing spin one-half atoms in a spin-up state ( I i > ).
Interposed in the beam we insert a second Stern-Gerlach apparatus
oriented at right angles so as to measure the spins in the left/right direction
(I -+ ) , I � ) ) , and the beam duly splits as required. However, if we try to
imagine the alternative input, corresponding to l l ) , we find that, instead of
being the alternative beam of the original Stem-Gerlach apparatus , it is yet
another beam , coming from the laboratory wall or some irrelevant part of
the apparatus ! Such a behaviour would normally be thought of as extremely
improbable, if not totally impossible - and not at all that given by the
quantum probabilities applied in the time-reversed sense.
An even more elementary situation of this type occurs if we consider a
photon , coming from a laboratory source (state Ii>), and simply reflect it off
a half-silvered mirror, where we set up photo-cells in both the reflected and
transmitted positions (states I -+ ) and I � ) , respectively). The state l l >
represents the photon coming from such a direction that if reflected i t would
reach the transmitted position and if transmitted it would reach the reflected
position . There would normally be no source in that direction. The photon ,
in state l l ) , like the atoms above, would have to be just 'absurdly' ejected
from the laboratory wall !
I believe this to be a key issue; and , since it is easily misunderstood , let me
try to be somewhat clearer about the sense in which these probabilities are to
be interpreted. It is necessary to adopt a viewpoint on this which allows
conditional probabilities to be understood in a way that is totally unbiassed
with respect to the direction in which time is taken to be running. Let me
simplify the above experimental set-up even further so that we have just one
photo-cell and one 'lamp' (the photon source). Half-way between the photo
cell and lamp we have a half-reflecting mirror whose plane is perpendicular
to the line joining photo-cell to lamp. If desired, we can add an ellipsoidal
mirror (a prolate ellipsoid of revolution) surrounding all three, the photo
cell P being at one focus and the lamp L being at the other, so that we need
not worry about the direction in which the photon enters or leaves the
photo-cell or lamp. (See Fig. 3. 1 .) There are four possible ways that the
photon can go. It can start at L and end at P, or it can start at L and end at L.
(These are the 'normal' cases , of everyday experience .) Moreover, it can start
at P and end at P, or it can start at P and end at L. (These cases are also
40 R. Penrose
needed in order to complete the space of quantum-mechanical transitions.)

The respective spacetime descriptions of these four possibilities are given in
Fig. 3 . 2 .
Now suppose that this 'experiment' is performed a very large number of
times at various different locations throughout spacetime (where, to be on
the safe side, I am restricting these locations to be in our own galaxy in the
present era). To see what the various conditional probabilities are, we
examine the spacetime (Fig . 3.3) an'd select those cases in which the in-state
is the one desired (I . . ) ) and count the various alternative out-states ( ( . . 1 )
(normal use of conditional probabilities) , or else we select those cases in
which the out-state is the one desired ( ( . . i) and count the various
alternative in-states (I . . ) ) (time-reversed use of conditional probabilities).
All the quantum-mechanical matrix elements (P I L) , (L I L) , (PI P) , (L I P>
have squared modulus equal to one-half. These give the correct probabilities
if we select for the in-state IL> and count how many times the photon ends
up in (Pl and how many times it ends up in (LI (closely 50 per cent of each) .
This is using the quantum-mechanical probabilities in the normal direction
in time. However, if we select for the out-state (LI and count how many
times the photon started out in IP) and how many times it started out in IL>
(closely 0 per cent and 100 per cent , respectively), we find that the reversed
direction quantum-mechanical calculation gives completely the wrong
answer (50 per cent for each) !
It is no good trying to retrieve the situation by bringing in other factors
that we do or do not know about the universe (e.g. the second law of
thermodynamics). We do not ask that other factors be taken into
consideration when calculating quantum-mechanical probabilities in the
normal direction in time. Quantum-mechanical probabilities are supposed
Fig. 3 . 1 . Is state-vector reduction time-symmetric? In this simple

experimental set-up a photo-cell P and a lamp L are separated by a half
silvered mirror. A photon travels between them.
Half-silvered
Ellipsoidal mirror
mirror
� - -
� -'.- ....· "' "'
--
t o be entirely stochastic, and not influenced by any other such factors. The
standard rules of quantum mechanics were obtained by observing (in our
own galaxy in the present era !) the way in which the probabilities behave in
the normal direction in time. These particular quantum-mechanical rules for
calculating probabilities simply do not work when used in the reverse
direction in time.
I am proposing to disallow these 'absurd' states (like the in-state jP)
above), and simply accept that the objective physical process underlying
state-vector reduction must be time-asymmetrical. (Most such 'absurd'
states, like the in-state IP) , would , when traced backwards in time,
ultimately becomes incompatible with the hypothesis of vanishing Weyl
Fig . 3 .2 . Spacetime diagrams of the four different ways that the photon can
travel . The top two diagrams depict the 'normal' situation, with the photon
coming from the lamp. The two lower possibilities are needed in order to
complete the list of possible transitions between the quantum states.
+
Time
I
p
p p
42 R. Penrose
cu rvature at initial singularities, since they are in contradiction with the

second law of thermodynamics.) Thus, one input can have several
alternative possible outcomes , but there would not generally be two
different inputs leading to the same outcome. This would mean that if we
follow a path in the phase space of an isolated system , then at certain stages,
namely where the state-vector reduction process objectively takes place, this
path would split into several paths, but different paths would not generally
unite together in this process. Thus , the evolution in phase space that is
implied by such a state-vector reduction process will be time-asymmetric :
paths can sometimes bifurcate but they do not rejoin .
3.1 3 Reduction and the (longitudinal) graviton count

Apart from merely this fact of time-asymmetry that I am claiming for the
reduction process, is there any more direct evidence to link this process with
gravity? I have argued elsewhere that there may well be (Penrose, 1 979,
198 1 , 1986). The quantum gravity effect that I claim must restrict past
singularities to have vanishing (or small) Weyl curvature also disallows white
holes (the time-reverses of black holes). For an isolated system of large
enough mass , this entails that paths in the phase space must sometimes
become united into a single path. This occurs because information can get
Fig. 3.3. The entire space-time is examined . If the initial state is selected for,
the final state ratios are compatible with the quantum-mechanical
calculation, but if the final state is selected for, the same calculation gives
quite the wrong answer for the initial state ratios.
< < t
<
Time
I
�
< <
<
<
�
lost at the singularities, all of which are of the future type (into which matter
can disappear : black holes) rather than of the past type (from which matter
could have emerged : white holes), the latter having to be effectively excluded
owing to the above-mentioned claimed quantum gravity effect. As I have
suggested elsewhere (Penrose, 198 1 , 1986), it is very plausible to suggest that
the uniting of phase-space paths that result from the above singularity
discussion is exactly balanced by the splitting of phase-space paths which I
argued earlier should result from the proposed objective state-vector
reduction procedure. I am not , of course, arguing for any direct connection
between specific spacetime singularities and specific instances of the
reduction process. Clearly one does not need a black hole in the laboratory
in order to perform a Stern-Gerlach experiment ! I am speaking merely of an
overall balance between phase-space loss from spacetime singularities and
phase-space gain arising from state-vector reduction. The argument is that if
the first is a quantum gravity effect, then so also must be the second.
In earlier accounts I have phrased things in terms of some (undefined)
concept of 'gravitational entropy'. My present opinion is that this concept is
both inappropriate and unnecessary for the discussion. (I am grateful to
Boris Zel'dovich and Bob Wald for some very pertinent criticisms of my
earlier view.) Instead, I believe what is required is some measure of
(longitudinal) graviton number. The idea is that when two states in linear
superposition become differently coupled to the gravitational field , to the
extent that the difference between the fields is of the general order of one
(longitudinal) graviton, then failu re of linear superposition becomes
significant , and some non-linear procedure (instability?) forces the state out
of the superposition and into one alternative or the other. This procedure, if
it could be found, would provide our sought-for objective state-vector
reduction .
The immediate question that arises is whether the orders of magnitude
involved make this seem at all plausible. While I have not been able to
explore this question in a great deal of detail, it seems to me that the signs
are not altogether discouraging. Let me give an example, which I shall try
to describe according to the kind of picture that I have in mind. Suppose we
have an atom surrounded, at some distance, by a particle detector - say a
cloud chamber or a bubble chamber. The atom decays, emitting a charged
particle. As the particle moves away from the atom, its reality consists of a
spherical wave moving outwards and centred on the atom. Quantum
linearity holds excellently at this level, and the wave can, if preferred, be
viewed as a linear superposition of straight particle tracks directed outwards
44 R. Penrose
from the atom. As these tracks enter the detector, they each cause streaks of
ioniz.ation which in turn cause trails of droplets, or bubbles, to be formed .
The reality o f the situation is still a complex linear combination o f all these
possible different trails together, so long as the movements of energy are
sufficiently small that the coupling to gravity is significantly below the level
of one graviton . Eventually, however, when the droplets, or bubbles , grow
to a sufficient size, the differences between the gravitational fields produced
by the different linearly superposed trails will grow to become quantum
mechanically significant - by which I mean that the (longitudinal) graviton
count for the differences between the various gravitational fields of the
individual trails reaches order unity. When this happens, non-linear effects
become important and the system 'flops' into a state in which the spacetime
geometry is (at the level of one graviton or so) well defined . Just one of the
complex-linearly superposed spacetime geometries is singled out, and so just
one of the possible trails reaches realiz.ation. The idea is that, in a sense,
Nature abhors complex linear combinations of differing spacetime
geometries!
One must do a calculation to see if this is in any way plausible. My own
crude attempts at this have been superseded by a calculation performed by
Abhay Ashtekar. Let us, for simplicity, assume that we have just a single
droplet. (The case of a bubble is essentially similar : we simply count its mass
as negative - which occurs in the formula squared. This mass is the difference
between the mass of the bubble and that of the ambient medium.) Let the
droplet's mass be m and its radius be r. Let the radius of the material from
which the droplet has condensed be R . Take everything to be spherically
symmetrical , and suppose, crudely, that there is a region of vacuum formed
between the radii R and r. Adapting, to the case of (linearized) gravity,
Zel'dovich's ( 1 966) procedure for estimating the number of photons in a
classical free electromagnetic field and applying the expression to the
situation of a static field with source (where the procedure does not strictly
apply - but what else can one do?), and adopting a particular gauge
condition , Ashtekar obtains the following expression for the expectation
value of the number of (longitudinal) gravitons arising in the region between
radii r and R (this being the region where the linearized Weyl curvature is
non-zero) :
768rc3(m/m p ) 2 log(R/r).
Here mp is the Planck mass ( 10 - 5 g). The coefficient 768rc3 is about

"'
twenty-four thousand, so effects at the 'one-graviton' level could show up

when m is around 10 - 7 g or so . The hope is that the corresponding formula

for a whole string of correlated droplets (or bubbles) , and other cases
without spherical symmetry, would lead to a somewhat smaller mass giving
a one-graviton effect.
The situations with other types of measurement, such as (developed?)
photographic plates, spark chambers, computer memories and , most
importantly , bundles of nerve signals (since the retina can be used as a
quantum detector for small numbers of photons - and because the only time
we know that reduction has taken place is when we personally experience it !)
remain to be examined . My impression is that it would be more encouraging
if plausible theoretical reasons could be found to suggest that the non-linear
reduction process should take effect at a somewhat smaller scale than is
estimated at present. Perhaps taking into account asymmetries, or detailed
granular structure, might lead to such a conclusion . Alternatively, it might
be that one needs a slightly different criterion for the onset of reduction, still
involving a 'graviton' criterion , but where there would be significant
probabilities for reduction over short timescales even well below the 'one
graviton' level . I feel that things are not too discouraging, in view of the
primitive nature of all these considerations, so far.
3. 14 Non-locality in quantum geometry

From the discussion given above, any physical reduction process must have
decidedly non-local and peculiar features. In this respect, it is encouraging
that the expression for graviton number in a linearized gravitational field is
indeed a non-local one. However, the reader should be warned, in my
opinion , that any successful objective physical theory of state-vector
reduction must be non-local in a way that fundamentally affects even the
very fabric of spacetime. I do not expect, except perhaps at some temporary
provisional level , that we shall be able to get away with a theory which
describes an objective reality taking place within some ambient spacetime.
The spacetime must itself become subject to this non-local description .
One reason for believing this stems from consideration of the kind of non
locality that arises in situations of the Einstein-Podolsky-Rosen type as
exhibited by the experiments of Freedman and Clauser, and Aspect and his
coworkers. We can obtain a consistent description of the photon pairs using
the conventional quantum-mechanical reduction procedure, but this must
be applied simultaneously over a region of space encompassing both
photons : a partial measurement is made on the system when one photon is
46 R. Penrose
observed, and this instantly affects the state of the other photon, putting that
one into a state which becomes subject to the observation on it which
immediately follows. However, this picture is not at all Lorentz invariant. If
we allow a different Lorentz frame to describe the situation, we can arrange
that it is the observation on this second photon which occurs first and causes
the partial reduction, the result of which 'then' becomes subject to the
original observation on the first photon. In this second Lorentz frame our
picture of 'reality' is completely different from the one in the first frame -
where I am taking the conventional description in terms of evolving and
reducing state-vectors to represent 'reality'.
At this point we could simply abandon relativity for our picture of reality,
if we wish. There would be no actual conflict with the results of experiment if
we do so . But there is conflict with the spirit of relativity in such a procedure
(cf. Bell , 1987). Once we adopt a J?.On-relativistic view for our picture of
reality, we find it to be a 'fluke' that relativity actually holds. To adopt such a
view would be to ignore the insight that Einstein impressed upon us in 1905
- and also the closely related insight that I am suggesting guided the
youthful Newton to an important aspect of his views on the nature of light.
This is the insight that relativity should be taken as a principle rather than as
an accidental feature of the precise form of the physical laws. It may be
argued that cosmology (and, indeed, the big-bang singularity) now supplies
us with a rest-frame, so the case is certainly debatable. Nevertheless, my own
money, for what it is worth, would go on a description of reality which, at
this level , should indeed be relativistically invariant. The best suggestion
that I can make at this stage would be for a picture involving some sort of
partially formed , partially bifurcating spacetime, where the nature of the
spacetime has not been adequately resolved until the second photon
observation has taken place. I have been suggesting , after all , that the
reduction process should be an intimate feature of the interrelation between
quantum mechanics and general relativity. General relativity describes the
structure of spacetime, so, when quantum mechanics becomes intimately
involved, we must expect some radically different pre-spacetime notions to
enter the picture.
Another puzzle concerning the reduction procedure is the one mentioned
above concerning 'null measurements'. Not observing a particle in a
detector can effect a partial reduction of the state-vecto r. On the view that I
am putting fo rward, this is not unreasonable. The spacetime geometry
which gets coupled to the situation resulting from the particle entering the
detector differs from the geometry resulting from it not entering the detector,
and their difference might well reach the one-graviton level. At this point
Nature would have to choose between (partial?) spacetime geometries. She
might well choose the geometry in which the particle has not entered the
detector - and that choice would non-locally affect the state elsewhere.
How is one to develop such a non-local quantum spacetime theory? I
have my own ideas about what might be the fruitful line to pursue, but it
would not be appropriate for me to try to enter into these here since the
ideas, even after very many years of labour, are still much too ill-formulated
to suggest any reliable insights. It will surprise no-one if I say that it is still
my opinion that the formalism of twistor theory, with its essentially non
local description of spacetime ideas, ought to provide an important input to
such a quantum spacetime theory (cf. Penrose and Rindler, 1986). Also , I
still have some hankering after ideas that I entertained many years ago in
evolving a theory of spin-networks (Penrose, 197 1 , 1972) . Thought
experiments of the Bohm-Einstein-Podolsky-Rosen type had an important
motivational role to play in that theory, and the idea was to build up the
concept of space as a limiting implicit structure when large numbers of
particles are involved . However, neither twistor theory nor spin-network
theory have any time-asymmetrical ingredients, as things stand . It is clear to
me that some essential new input is needed.
Whatever the future may hold for the development of physical theory, it is
clear that quantum theory, relativity, and even Maxwell's electromagnetic
theory, have led us - or should have led us, if we have examined the facts - to
a view of reality very different from that presented by classical Newtonian
particle mechanics. From his writings, it is equally clear that Isaac Newton
was not, in the strict sense, a 'classical Newtonian'. His view of Nature
contained very much more mystery than that which the standard picture of
Newtonian mechanics now conjures up. Perhaps his readiness to put
forward such a strange corpuscular yet wave-like conception of light is
indicative of his belief in the depths of Nature's mysteries. If so, he was
certainly right !
Acknowledgements
I am grateful to Abhay Ashtekar and Dipankar Home for valuable
suggestions, and to Karel Kuchar for very helpful criticisms of the
manuscript . I thank, also, the Institute for Theoretical Physics at the
University of California in Santa Barbara for its hospitality and stimulation,
where discussions with D. Boulware, I. Bialynicki-Birula, E. T. Newman,
W. G. Unruh and R. Wald were most valuable. This research was supported
48 R. Penrose
in part by the National Science Foundation under Grant No. PHY82-

1 7853, supplemented by funds from the National Aeronautics and Space
Administration, at the University of California at Santa Barbar:a and in part
by the U.S . Air Force Office of Scientific Research.
References
Arnol'd , V . I . ( 1 978). Mathematical Methods of Classical Mechanics. Springer : New

York.
Aspect, A. ( 1 976). Phys. Rev. D l4 , 1944.
Bell, J. ( 1987). Beables for Quantum Field Theory in Quantum Implications: essays in
honour of D. Bohm (eds. B. J. H iley & S. D. Peat, Routledge & Kegan Paul) p. 227
Drake, S. ( 1957) . Discoveries and Opinions of Galileo. Doubleday : New York .
Freedman, S . J. and Clauser, J . F. ( 1972). Phys. Rev. Lett., 28, 938.
Galilei, G . ( 1638). Dialogues Concerning Two New Sciences (Macmillan edn , 19 14) .
Dover : New York.
Karolyhazy, F. ( 1966). Nuovo Cim., A42, 390.
Karolyhazy, F., Frenkel, A. and Lukas, B. ( 1 986). On the possible role of gravity in the
·
reduction of the wave function . In Quantum concepts in Space and Time, ed. C. J.
Isham and R . Penrose. Oxford University Press: Oxford .
Komar, A. B . ( 1969). Int. J. Theor. Phys., 2 , 257.
Kuchar, K . ( 1 98 1 ). Canonical Methods of Quantization in Quantum Gravity 2: a
second Oxford Symposium (ed. C. J . Isham, R. Penrose & D. W. Sciama, Clarendon
Press, Oxford) p. )29
Mache, E. ( 1 972) . The Science of Mechanics. Open Court : La Salle, Ill.
Newton, I. ( 1730). Opticks. ( 1952) Dover : New York.
Newton, I. ( 1687). Principia.
Pais, A. ( 1980). Subtle is the Lord. Oxford University Press : Oxford .
Penrose, R . ( 197 1). Angular momentum : an approach t o combinatorial space-time. I n
Quantum Theory and Beyond, ed . T. Bastin . Cambridge University Press : Cambridge.
Penrose, R. ( 1 972). On the nature of quantum gravity. In Magic Without Magic, ed.
J. R . Klauder. Freeman : San Francisco.
Penrose, R. ( 1 976). Gen. Rel. Grav., 7, 3 1-52.
Penrose, R. ( 1978). Singularities of space-time. In Theoretical Principles in Astrophysics
and Relativity, ed. N . R . Leibowitz, W. H . Reid and P . 0. Vandervoort. Chicago
University Press : Chicago.
Penrose, R. ( 1 979). Singularities and time-asymmetry. In General Relativity: An Einstein
Centenary, ed. S . W. Hawking and W. Israel. Cambridge U niversity Press,
Cambridge.
Penrose, R. ( 198 1 ). Time-asymmetry and quantum gravity. In Quantum Gravity 2, ed .
C. J. Isham, R. Penrose and D. W . Sciama. Oxford University Press : Oxford .
Penrose, R . ( 1 983). Donaldson's moduli space : a 'model' fo r quantum gravity? In
Quantum Theory of Gravity, ed. S . M . Christensen . Adam Hilger: Bristol.
Penrose, R. ( 1986). Gravity and state-vector reduction. In Quantum Concepts in Space
and Time, ed. C. J. Isham and R. Penrose. Oxford University Press : Oxford .
Penrose, R. and Rindler, W. ( 1984). Spinors and Space- Time, Volume 1 . Cambridge
University Press : Cambridge.
Penrose, R . and Rindler, W. ( 1986). Spinors and Space- Time, Volume 2 . Cambridge
Westfall, R. S. ( 1980). Never at Rest. Cambridge University Press : Cambridge.
Wheeler, J. A. ( 1964). In Relativity, Groups and Topology: the 1 963 Les Houches
Lectures, ed . B. S. De Witt and C. M . De Witt. Gordon and Breach : New York.
Wheeler, J. A. (1 964). Geometrodynamics and the Issue of the Final State in Relativite,
Groupes et Topologie (eds. C. deWitt & B. deWitt, Gordon and Breach, New York &
London)
Wigner, E. P. ( 1961). Remarks on the mind-body question. In The Scientist Speculates,
ed . I . J. Good . Heinemann : London .
Zel'dovich, Ya. B . ( 1 966). Sov. Phys. Dok/., 10, 77 1-2 .
Note added in proof

It has been brought to my attentio n , by N . M . J. Woodhouse, that in
Newton's manuscript fragment De motu corporum in mediis regulariter
cedentibus (a precursor of Principia written in 1684) Newton had originally
proposed to base his mechanics on five (or six) fundamental Jaws rather than
the three that have come down to us through Principia . Law 4 was actually a
clear statement of the (Galilean) relativity principle! Newton was well aware
that these Jaws were not independent of one another, and for Principia he
settled on the three that are now familiar to us. What is remarkable -
apparently supporting the viewpoint tentatively put forward in this article -
is that Newton had at one time, indeed, seriously contemplated using the
relativity principle as a fundamental principle (despite what his views on
'absolute space' may or may not have been)!
4
Experiments on gravitation
A.H.COOK
We at least know something in knowing what qualities gravitation does

not possess and when the time shall come for explanation, all these
laborious and, at first sight, useless experiments will take their place in
the foundation on which that explanation will be built.
J. H. Poynting ( 1900)
4. 1 Introduction
Newton's law of gravitation has long intrigued and balled experimenters
and theorists alike by its simplicity and generality and by the failure of all
attempts to establish any departure from it or any dependence on
extraneous circumstances. The only known deviation is that consequent
upon general·relativity in the neighbourhood of a massive body which leads,
for example, to the anomalous precession of the perihelion of Mercury. The
situation is in contrast to that of the force between electrically charged
particles where the inverse square law attraction between isolated stationary
particles is modified by the presence of other material bodies and by the
velocities of particles. Those deviations from the simple Coulomb law have
enabled the nature of the electromagnetic force and the electrical structure of
materials to be studied by experiment and , in the absence of similar effects in
gravitation, the gravitational force resists experimental investigation .
It may be that gravitation truly is nothing but a manifestation of the
geometrical structure of nature as set out in general relativity, but it may be
that the reason that no other effects have so far been detected is that they are
very small and that experiments are insufficiently sensitive. As will appear
later on , experiments on gravitation are indeed much more difficult than
those on electromagnetism , for two reasons : first, the gravitational force is
only 10 - 4 0 of the electrostatic force between baryons, so that disturbing
forces are much more troublesome; and , secondly, the techniques available
for mechanical measurements are far less delicate than those for electrical
measurements, in which very sensitive electronic devices can be used. Thus,
for example, the inverse square law variation of gravitation has been shown
Experiments on gravitation 51
to be correct to within about 1 part in 104 at distances of 0. 1 m, whereas the

limits set by experiment on the inverse square law of electrostatic attraction
are many orders of magnitude less. Again , the constant of gravitation is
known to little better than 1 part in 104 whereas almost all other
fundamental constants of physics are known to 1 part in 106 and some very
much better. Experiments on gravitation are therefore a challenge to the
technique and imagination of the experimenter and so continue to attract
skilled and ingenious physicists.
Experiments on gravitation are complementary to analyses of celestial
mechanics. It is well known that despite the most thorough attempts
(including some with notorious errors in dynamical theory) no departure
that can unambiguously be attributed to a deviation from the inverse square
law has ever been established, with the sole exception of the consequences of
general relativity. In particular, the inverse square law itself is firmly
established in celestial mechanics and the force of gravitation shows no
dependence on the direction of the line joining masses. These two properties
can indeed be investigated in the laboratory at short distances, but with very
much greater uncertainties. On the other hand , celestial mechanics tells
nothing about the dependence of the force on the constitution of the masses
(nor indeed that it is just proportional to the product of the masses) whereas
that is something that can be studied with great precision in the laboratory.
Something can also be learned from geophysics, at distances intermediate
betw�en those of the laboratory and celestial mechanics, and limits may be
placed on directional effects and on shielding. Celestial mechanics,
geophysics and laboratory investigations must therefore all be taken into
account in assessing the validity of Newton's law, but this article is
concerned almost entirely with experiment, that is to say, with observations
which can be made in a laboratory where conditions are under the control of
the experimenter. Nothing is therefore said about the laser ranging
observations of the Moon, which belong to celestial mechanics. Recently it
has become possible to think of making experiments in space craft and that
has somewhat extended the range of effects that may be studied - however,
no such experiment has yet been carried out.
Experiments on gravitation have been made from the beginning. Newton
himself, no mean experimenter, used pendulums of different materials to see
if there were any dependence on constitution and was able to set a limit of
about 1 part in 103 (Newton, 1687). He also made some estimates of the
gravitational attraction of a mountain on a plumb bob and of the time it
would take two masses to come together under their mutual gravitation,
52 A. H . Cook
both with the idea of determining the constant of gravitation. He concluded

that the effects were too small to observe but the first was employed by
Bouguer and by Maskelyne ( 1 775) (at Schiehallion) while the second was the
principle of an experiment attempted some twenty years ago (Luther et al.,
1976). The most important ad vance in experiments on gravitation and other
delicate measurements was the introduction of the torsion balance by
Michell and its use by Cavendish ( 1 798) in gravitation and electrostatics .
It has been the basis of all the most significant experiments on gravitation
ever since. It is also a measure of the inherent difficulty of gravitational
experiments that, despite many refinements of technique and design, the
accuracy of Cavendish's determination of G has been little improved upon
until very recently. By contrast , the study of possible differences between
materials has been carried to great refinement, with limits now about 1 part
in 10 1 2 instead of the 1 in 10 3 that Newton attained . These and other
experiments of historical interest will be described to some extent in the
appropriate sections.
4.2 Theoretical framework

This section is in no sense a technical account of theory. There are indeed
two approaches that may be taken to the theoretical grounds for
experiments. One is to explore the inferences to be drawn from a consistent
theory of gravitation and to see what experiments may be conducted to
check them. Will ( 1 98 1) has shown that many theories may be treated on a
uniform basis in the parametrised post-Newtonian (PPN) formulation and
it is convenient to follow him in setting out the implications of particular
experiments. On the other hand, one could, in a rather simple-minded way ,
just take Newton's law and ask what features of it might be tested
experimentally : is force just proportional to the product of masses, are non
linear or shielding effects absent? is the force inverse square? does it depend
on direction? on the constitution of the masses? on time? on velocities? It is
convenient to combine the two approaches - the type of experiment that can
be done is best classified according to the second scheme, the interpretation
is conveniently given in PPN terms.
The inverse square law itself is somewhat special because of a direct
connection with the inverse square law for electrostatics, the validity of
which has been established experimentally to very high precision .
The conditions for the inverse square law to be followed by a force are that
the geometry of space should be locally Euclidean and that the mass of the
force field should be zero. The first condition is equivalent to the area of the
sphere drawn about a point source being proportional to the square of the
radius, or to the proper distance from the source being independent of
source strength , for Gauss's theorem shows that in a massless field the
sou rce strength is equal to the integral of the force over a sphere drawn
around the source and, if the inverse square law is to apply, the area of the
sphere must be propo rtional to the square of the radius. In the PPN
formulation , the proper distance from a source of potential U is r( l + y U)
where y is one of the PPN parameters (see below) ; if y is zero, the proper
distance is independent of U and then the area of a sphere is proportional to
the square of the radius.
The condition that the field should be massless is not satisfied by, for
example, a Yukawa potential ; effectively the total energy stored within a
sphere now depends on the size of the sphere, even with Euclidean geometry.
The electromagnetic field is massless as is shown by the fact that
electromagnetic waves in vacuum have no dispersion . The electrostatic law
of force is also inverse square to very close limits, as shown experimentally
by the absence of any detectable field outside a closed conductor. It may
therefore be inferred that the geometry is locally Euclidean, at least for the
values of masses that can be placed in a terrestrial laboratory.
If then , it were found that the inverse square law did not hold for
gravitation, it would follow that the gravitational field carried mass, and the
analysis of experiments on the inverse square law is then naturally made in
terms of a Yukawa potential (Fujii , 197 1 , 197 2 )
k
- - ( 1 + ex e - µ r).
r
Experiments on the inverse square law seek to establish limits on ex
and µ.
The general structure of the PPN formulation of metric theories of
gravitation has been described by Will ( 198 1). The theories are metric
because the motion of a particle is completely determined by the metric of
the geometry, a metric which is specified by ten parameters.
The elements of the metric form are
2
900 = - 1 + 2 U - 2 p u - 2e<1>w + (2y + 2 + ex3 + ( 1 - 2 e ) <1>1
+ 1(3y - 2 p + 1 + c 2 + e ) <1> 2 + 20 + ( 3)<1>3 + 2(3y + 3(4 - 2e )<1>4

2
- (( 1 - 2e ) d - (ex 1 - ex 2 - ex3)w U -ex 2 w w Uii + (2ex3 - ex 1 )w i Jti;
i i
9 oi = -!(4y + 3 + ex1 - ex 2 - 2e ) Vi - -!{ l + cx 2 - C1 + 2e ) �

i
--!(ex 1 - 2ex 2 )w U - ex 2 wi Uii ;
9ii = ( 1 + 2y U) c5ii·
54 A. H. Cook
where U is the Newtonian scalar potential, <l>w is a potential introduced by

Whitehead, <1> 1 , <1>2 , <1> 3 , <1>4 and ,9/ are other scalars, J,'{ and W; are vector
potentials and Vii is a tensor potential . In the absence of sources, the metric
coform is that of special relativity, diag ( - 1 , 1 , 1 , 1 ) wi is the velocity of the
.
local frame relative to some absolute frame.

In general relativity, f3 = y = 1 and all other parameters are zero , and so the
metric is
g 00 = - 1 + 2 U - 2 U 2 + 2(2 <1> 1 + 4<1> 2 + <1> 3 + 3 <1>4 ) ;
9o i = - -! J-'i - ! W;;
g ii = (1 + 2 U)b ii '
where <1> 1 , <1> 2 and <1> 3 are scalar potentials of which the sources are kinetic,
potential and internal energy not included in the rest masses. Most of the
experiments that are considered below will be related to the PPN
parameters.
A crucial postulate of general relativity is the weak equivalence principle,
according to which the passive gravitational mass of a test particle bears
always the same ratio to its inertial mass , whatever its material constitution .
The motions of all test particles in a gravitational field are therefore all the
same, in fact . along geodesics. Thus all bodies in vacuum fall towards the
Earth with the same velocity. If this were not so in celestial · mechanics,
Kepler's law,
relating the (active) mass M of a primary to the semi-major axis, a , and mean
motion , n, of a secondary would not be followed, for it assumes that the
inertial mass, mi , and (passive) gravitational mass, mg , of the secondary are
the same. Without that assumption,
GM = a 3 n 2 (mg/mi )
and if secondaries differed in the ratio mg /mi (as , say , between the Earth and
Jupiter) there might be detectable differences.
The weak equivalence principle does not hold in all PPN theories, for the
ratios of passive and active gravitational masses (mp and m A ) to inertial
mass, mi, are given by (Will , 198 1)
mp /mi = 1 + (4{3 - y - 3 - 13° e - cx1 + jcx2 -1(1 - !(2 )n/m
and
mA jmi = 1 + (4 {3 - y - 3 -.!fe -!cx 3 - !( 1 - 2 (2 ) Q/m
+ ( 3 E' /m' - (-!cx 3 + ( 1 - 3( 4 )P' /m' ,
where m i is the total mass energy of the body (rest energy, internal kinetic,
potential and gravitational energy) , n is the internal self-energy of
gravitation , E' is the integral of internal energy and P' that of pressure for the
attracted body of mass m' . In general relativity 4P = y + 3 and all other
parameters are zero , so mA = mp = mi . Experimental studies of the weak
equivalence principle are crucial to gravitational physics.
The crucial observations that first established the validity of general
relativity were those on the deflexion of light in the gravitational field of the
Sun and the anomalous advance of the perihelion of Mercury,
corresponding to a term in the potential of the Sun proportional to 1/r 2 . The
deflexion of light may be seen as the consequence of a reduction of the speed
oflight near a massive body ; the reduction , which is proportional to 1( 1 + y),
may be observed either through the shift of the apparent direction of a
sou rce or through an increase in the time of passage of light. Both
observations can now be made rather precisely by radio means, long-base
line interferometry for direction (Fomalont and Sramek, 1977), and radar
measurements (Shapiro et al. , 1972) and radio transponders (Anderson et
al., 197 5 ; Reasenberg et al., 1979) for time delays. The time delay
measurements are the most precise and show that y is 1 to within 0. 1 per cent
(Will , 198 1 ) in accordance with general relativity. The overall precession of
the perihelion of Mercury is obtained both from long series of optical
observations (to within 1 per cent) and from radar observations of the
distance of Mercury from the Earth (to within 0.5 per cent) (Shapiro et al.,
1972 ; Shapiro et al., 1976) . In Newtonian theory, the attractions of the other
planets contribute to the precession , as does the quadrupole moment of the
Sun and there is an effect of the general precession of the equinox . The
relativistic effect is the residue. The planetary and equinoctial terms are well
known, but the value of the quadrupole moment of the Sun has been
questioned . The relativistic precession is proportional to �(2 + 2y - p) and
with the best radar results for the overall precession and discounting doubts
about the internal constitution of the Sun, p is found to be 0.99 1 + 0.0 15,
again in accordance with general relativity.
It thus appears that those PPN parameters, P and y, which are non-zero in
general relativity, have the expected values of 1 . There is also nothing in the
motions of planets or satellites which requires there to be a preferred frame
of reference for the solar system. In the PPN formulation, the equations of
motion contain terms proportional to the velocity relative to the preferred
frame, multiplied by various linear combinations of the parameters r.< 1 , r.< 2
and rt 3 . The absence of preferred frame effects means that r.< 1 , r.< 2 and r.< 3 are all
zero. Evidence from Earth tides (Section 4.5) also sets limits on the r.<s.
56 A. H . Cook
4.3 The inverse square law

In the hands of Cavendish, the torsion balance was used to establish the
inverse square law for electrostatic forces. Subsequently, the inverse square
law in electrostatics has been demonstrated to the highest precision by
showing that no field can be d etected outside a closed conductor containing
charge (the Faraday cage). Experiments on gravitation are far more difficult
and it is only with the refinement of the torsion balance in recent years that
useful results have been obtained, Cavendish ( 1 798), in his initial
determination of the constant of gravitation, mentions that he has made a
check of the inverse square law but gives no details, and in subsequent
determinations (for example, Baily, 1842 ; Cornu and Baille, 1873, 1878) no
check on the law seems to have been attempted , but the overall consistency
of results from experiments in which the distances between masses were
different, implies a rough agreement with the inverse square law. Mackenzie
( 1895) in his experiments on the attractions of crystals, placed the masses at
separations of 3 .6 , 5 . 5 and 7.4 cm and claimed that he had confirmed the
inverse square law to 1/500 between 3 . 5 and 7.3 cm.
C. V. Boys ( 1 895) greatly increased the sensitivity and stability of the
torsion balance when he made very fine quartz fibres and , subsequently,
tungsten wires properly treated (Braginski and Manukin , 1977 ; Chen ,
Cook and Metherell , 1984) have proved as satisfactory. It is not sufficient
that a fibre should have a small restoring couple; more important, its rest
point should be stable. The other improvement has been in the means of
detecting the angular position of the beam of a torsion balance ; the optical
lever used by Boys has been replaced by optical interferometers, capacitance
bridges or position-sensitive photo-detectors, each of which is more
convenient and capable of much greater sensitivity than the traditional
optical lever, to such an extent that it is not necessary to use the highest
sensitivity of which a torsion balance is capable, allowing more attention to
be paid to stability.
In recent years three experiments on the inverse square law have been
completed , in two of which forces between masses at different distances from
a test mass were compared while, in the third , a gravitational equivalent of a
Faraday cage was used . The interpretation of all such experiments depends
on being able to calculate the forces corresponding to a specific non
N ewtonian potential , in particular one in the form
- (k/r)( 1 + � e - µr).
The form
G = G 0 [ 1 + a ln(r/r 0 )]
has also been used for the non-Newtonian force ; the two are equivalent at
short distances if
£r 0- 1 rxµ .
=
The necessary calculations have been thoroughly investigated by Chen

( 1 982) (also Cook and Chen , 1 982) particularly for the cylindrical masses
that he used in his own experiment (Chen , Cook and Metherell, 1984).
Long ( 1976) appears to have been the first after Mackenzie to have
investigated the inverse square law at small distances . He placed two objects
of different mass in tum opposite a test mass on one end of the beam of
a torsion balance, at such distances that the forces on the test mass should be
equal for Newtonian attractions. His attracting objects were rings of
different radii . If the radius of a ring is r and the mass of the ring is m, then the
axial attraction at a distance z from the plane of the ring is
G2nmz
(r 2 + 2 2 ) 3 / 2 ·
The attraction is a maximum when z 2 equals 1r 2 , when it has the value
4nGm/3 3 1 2r2 .
If the rings are placed so that the attractions they exert upon the test masses
are maximum , it is not necessary to measure the (possibly ill-defined)
distances from the test mass . Long claimed tha� his results showed
significant deviations from the inverse square law (see below).
The principle of the experiment of Newman and his colleagues (Spero et
al. , 1980) was different : the field at the centre of an infinitely long hollow
cylinder would be constant if the inverse square law applied . Such a cylinder
is the gravitational equivalent of the Faraday cage and a test mass inside it
would experience no change of force as the cylinder was moved from side to
side. The test mass must be suspended from the arm of a torsion balance
used to detect changes of force, and so the cylinder cannot be infinitely long
and must be open at the ends. There will then be end effects such that the
mass will experience some residual variable force under the inverse square
law , but it is sufficient to calculate that from the Newtonian law. Spero et al .
( 1980) found no significant deviation from the inverse square law but Long
( 1980) criticised the principle of their experiment on the grounds that it did
not exclude all possible non-Newtonian effects and Chen (Cook , Chen and
Metherell , 1984) has performed an experiment the design of which meets
Long 's criticism and which has set slightly different limits to deviations from
the inverse square law from those of Spero et al. ( 1980).
58 A. H . Cook
Like Long, Chen compared the effects of d.ifferent masses on a test

mass, but his masses were quite large cylinders and he used three ; one was
placed in a fixed position relative to the test mass and its attraction was
balanced either by a small cylinder close to the test mass or a larger one
further away. Thus the test mass was always observed in the same position ,
or nearly so. In consequence, it is not necessary to find the position of the
test mass relative to the cylinders. Let m1 , m2 , m 3 be the masses of the three
cylinders, of which 1 is kept fixed . Likewise, let r1 , r2 and r 3 be the distances
from the test mass. Then it is difficult to measure those distances from the
mobile mass but straightforward to measure (r1 + r2), (r1 + r 3) and (r2 - r 3) .
With the Newtonian law,
2 2 2
r 1 = (r1 + r 2)m V /(m � 1 + m1f )
and
2 2 2
r 3 = ( r1 + r 3)m V /(m �1 + m V ) ,
whence (r2 - r 3) a s calculated from the masses and the measured sums
(r1 + r2) and (r1 + r3), may be compared with the observed ( r2 - r 3) for the
respective conditions of balance for masses 2 or 3 with 1 . Chen placed his
torsion balance in an evacuated enclosure and observed changes of position
by reflecting a beam of laser light from a mirror on the beam to a position
sensitive photo-detector.
The centres of mass of large objects used in experiments on gravitation
cannot be found exactly because the density may not be uniform but the
uncertainty may be eliminated by rotating a mass such as a cylinder about
its geometrical axis.
Chen in fact carried out two groups of experiment, in the one the net force
on the test mass was very close to zero , while in the other there was a
significant net force. The purpose was to check the existence of a vacuum
mass polarisation in the presence of a net field .
The results of experiments on the inverse square law may be expressed as
permitted ranges of the parameters a and µ. According to Fujii ( 197 1 , 1 972)�
a should be ! in which case the following limits are placed on µ :
Long ( 1 976) : (non-null) 1 .8 > µ - 1 > 2.2 m (Chen , l

1
Spero et al. ( 1980) : (null) µ - > 3 .4 m (Chen , 1 982)
Chen, Cook and Metherell ( 1984) : (non-null) µ - 1 > 3 .4 m
(null) µ l > 4.9 m
-
Alternatively, the results may be expressed as values of t: in the law G =

G 0 ( 1 + t: ln (r/r 0 ))
Long ( 1976) : (20 + 4) x 10 - 4
Spero et al. ( 1 980) : ( l + ) x 10 - 5

Chen , Cook and Metherell ( 1984) : (non-null) ( - 4.3 + 5. 1) x 10 - 4
(null) (2.0 + 6.6) x 10 - 5
Spero et al. ( 1980) have shown how the result of an inverse square law
experiment may be displayed in a diagram of allowed combinations of a and
µ and the diagrams for the Long, Spero and Chen experiments are
reproduced in Fig. 4. 1 . A summary of the results is that the inverse square
law is followed to within 1 part in 10 - 4 at distances of about 0. 1 m and that
the characteristic length, µ - 1 in a Fujii-type potential is at least 3 . 5 m (see
also, Gibbons and Whiting , 1 98 1).
Other laboratory experiments have been performed in recent years but
Fig. 4 . 1 . Allowed values of o: and µ in a Spero diagram (after Chen, Cook

and Metherell , 1984).
,/
�ll
Excluded
/I Chen et al.
Kn-null
1
//
I
Allowed
-4
-2 0 +2 lg (µ- 1 /m)
0
\ Excluded
I
I
I
/I
'/
�
I
/;
,....,
/;
I
�
-2 Allowed
..!!!l
'-'
/ -null
//
� Chen et al.
, ../7 Null
I
-4 \ /
-2 0 + 2 lg (µ-1 /m)
60 A. H . Cook
they are less precise than the three just described (Panov and Frontov, 1979;
Ogawa et al. , 1982 ; Chan et al., 1982). Experiments at greater distances have
also been attempted (Yu et al. , 1979) while Stacey and Tuck ( 198 1) have
argued from geophysical measurements that the corresponding value of G is
some 0.5- 1 .0 per cent greater than laboratory determinations. However, the
precision of these results also is poor (see also next section).
4.4 The weak equivalence principle

As already mentioned, Newton himself did some experiments with
pendulums of different materials. He made two pendulums with equal
wooden containers hung from wires 1 1 ft long , and compared the periods
when one container was filled with wood and the other with gold , silver,
lead, glass, sand, salt or water. He found the periods to be the same to about
1 part in 1000 (Newton , 1687), and Bessel ( 1 832) in a more thorough study of
12 different materials was able to set a similar limit. The big advance in
experiments in this field was made by Eotvos who , in the course of
developing a torsion balance for investigating the gradients of the local
gravitational field, saw how to use the torsion balance to compare the
attractions o ( the Earth and of the Sun upon masses of different
composition. Experiments of this sort in fact test for differences in the ratio
of gravitational to inertial mass, for some way has to be found of establishing
when masses are equal , independently of the gravitational forces upon them.
In pendulum experiments, where the period is equal to
2n(mggh/mi r 2 ) 112
(mg is the gravitational mass, mi the inertial mass, and r the radius of
gyration) it is clear that differences of period would correspond to different
ratios (mg /mJ The geometry of the Dicke and Braginski experiments is
clearer than that of the terrestrial Eotvos experiment, so it is convenient to
consider the former first.
In the experiments of Roll , Krotkov and Dicke ( 1964) and Braginski and
Panov ( 1972) the attractions of the Sun upon different masses are compared
with the inertial accelerations. Roll , Krotkov and Dicke used a torsion
pendulum with three masses on a triangular frame - two were aluminium
and one was gold , as shown in Fig. 4.2. Let M 0 be the mass of the Sun , R its
distance from the apparatus and w the angular velocity of the Earth about
its axis of rotation. Let mi be the gravitational mass of the ith body and m� its
inertial mass. Let Pi be the perpendicular distance from the point of
suspension to the line from the Sun to the ith body, projected upon the plane
of the triangle of mass.
The gravitational torque about the suspension fibre is then
GM O
(m 1 P 1 + m 2 P 2 + m 3 p 3 )
Ri ·
while the inertial acceleration is

Rw 2 (m�p i + m'2 P 2 + m'3 p 3 ).
For the Earth as a whole R , the orbital distance from the Sun to the centre of
mass of the Earth, adjusts itself so that GM 0 /R 2 is equal to Rw 2 (Kepler's
law).
If the ratios of gravitational to inertial mass are the same for each of the
three masses on the pendulum then the ratio
(m 1 P 1 + m 2 P 2 + m 3 p 3 )/(m� P 1 + m� P 2 + m'3 p 3 )
will be unchanged when the P i vary as the direction of the Sun changes with
the rotation of the Earth , but if the ratios do differ between materials then
Fig. 4.2. The Dicke experiment on the weak equivalence principle.
Al
Sun
62 A . H . Cook
the ratio of torques will not be constant, but will vary with a period of 24 h.
Roll , Krotkov and Dicke combined a very sensitive detector of the angular
position of the triangular frame with electrodes fed with a voltage to apply a
restoring force, so that the angular position of the frame was kept constant
as the Earth rotated ; a record was made of the voltage that had to be
applied.
One of the main problems that had to be faced was that many disturbing
effects depend on the position of the Sun, whether directly through the
diurnal variation of the temperature or otherwise, so that very great care
had to be taken to isolate the apparatus (in a special vault) to reduce
extraneous effects of diurnal period so far as possible. The limit put upon any
difference of the ratio of gravitational to inertial mass between gold and
aluminium was 1 part in 101 1 , but an unexplained feature of the
observations is that there was a variation of torque with a semi-diurnal
period . Subsequently Braginski and Panov ( 1972) carried out a similar
experiment but with improved discrimination, so that they were able to set a
limit of 1 part in 101 2 on any difference between materials.
Another experiment was started about 1979 (Keiser and Faller, 1979) in
which the test masses are not hung from a fibre but are floated on water at
the temperature of maximum density and are centred by electrostatic forces.
Preliminary results were encouraging but a final result has · still to be
announced.
EOtvos's main experiments were different from the later ones, in that
the gravitational attraction was that of the Earth and the inertial force that
of rotation about the Earth's polar axis. The geometry is shown in Fig. 4.3. The
forces acting on the masses are g, the gravitational attraction of the Earth
which is towards the centre of mass, and a, the centrifugal force which is
perpendicular to the axis of rotation. The resultant of the two is the force of
gravity as commonly understood and the suspension fibre is tangential to
the resultant ; that direction is not the direction of gravitational attraction
unless the centrifugal force is parallel to it (on the equator) or is zero (at the
poles).
Let IX be the angle between the gravitational and resultant forces and let ()
be that between the centrifugal and resultant forces ; () is very nearly 1 80°
minus the latitude. If the beam of the torsion balance hangs east-west, the
net torque on one mass is
gmg I sin IX ami I cos () sin (),
-
where I is the arm of the beam, mg is the gravitational mass and mi is the inertial
mass. With two masses, one at each end of the beam , the net torque is
g (m 1 g l 1 - m 2g l 2 ) sin cx - -!a (m 1 J 1 - m2 J 2 ) sin 28.
If the apparatus is turned round through 180° the signs of the differences
(m 1 g l 1 - m 2g l 2 ) and (m 1 J 1 - m 2 J 2 ) are reversed . If the two factors are equal
there is no change of torque on reversing the balance, but if they differ
because the ratios (mg /mi ) 1 and ( mg /mJ 2 are unequal , there will be a net
change of torque.
Eotvos, Pekar and Fekete ( 1922) examined a number of pairs of different
substances and found no difference of gravitational mass relative to inertial
mass to exceed 1 part in 109 and their results were confirmed in a later set of
experiments by Renner ( 1935).
Recently, the results have been reexamined by Fishbach et al. ( 1986).
Fig. 4.3. The Eotvos experiment on the weak equivalence principle.
North
/
/
/
/
/
/
/
/
/
/
/
/
/
g a
g +a
a a
g g
64 A. H . Cook
They point out that if there is an additional non-Newtonian potential of the

form
then the constants in that expression might depend on the material

constitution and in particular on the baryon numbers of the substances.
Fishbach et al. also appeal to certain geophysical evidence which may
indicate that the effective constant of gravitation at distances of some
hundreds of metres is greater than that for laboratory distances (Stacey and
Tuck, 198 1 ; Holding and Tuck, 1984) and would seem to entail a value of ex
of ( - 7 .2 + 3 .6) x 10 - 3 and of µ 1 of (200 + 50) m. Such values could not be
-
detected by the inverse square law experiments of Section 4.3. They would
also not affect the experiment of Roll, Krotkov and Dicke ( 1964) , nor of
Braginski and Panov ( 1972) for there r is the distance to the Sun and e µr is
-
negligible. There might , Fishbach et al. argue, be some effect in the original
Eotvos experiment, because in it a substantial part of the attraction on the
test masses, a few parts in 10 3 , comes from terrestrial matter within a few
hundred metres of the apparatus. Now the baryon number of material of
given mass varies with atomic mass on account of different binding energies,
and so a failure of the weak equivalence principle might be expected in the
circumstances of the Eotvos experiment. Fishbach et al. argue that such a
failure is indeed shown if the results are appropriately analysed, but Keyser,
Niebauer and Faller ( 1986) maintain that the analysis is faulty, that the
results of the further Eotvos type of experiment by Renner ( 1935) are
incorrectly neglected, and that the discrepancy as isolated by Fishbach et al.
is inconsistent with that needed to explain the geophysical data. The status
of the suggestion of Fishbach et al. is thus doubtful .
It should be noted that the force postulated by Fishbach et al. , involving a
new type of hypercharge, is different from that proposed by Fujii ( 197 1 ,
1972) which is purely gravitational.
The implications of the weak equivalence results are rather far-reaching.
Thus Dicke has pointed out that , since the relative numbers of protons and
neutrons differ in aluminium and gold , there is no detectable difference
between the attraction of the Sun upon protons and neutrons. Further, the
internal velocities at a given temperature are different because the atomic
masses are different and so the gravitational force cannot depend on atomic
velocities. (The kinetic energies are of course the same, so the experiments
say nothing about a potential proportional to kinetic energy.)
Experiments on gravitation 6S
4.5 Preferred frames and locations

In the PPN formulation the velocity of a local frame relative to some
preferred celestial frame enters the metric through the following terms :
. . .
m 9 00 : - (C( 1 - C( 2 - C( 3 )w 2 U - C( 2 w ' w1 Uii + (2 a 3 - a i )w' v; ,
•
. .
m g 0 i : - 21 (C( 1 - 2C( )w ' U - a w1 U ii '
.
2 2
and as a result (Will, 198 1) the measured constant of gravitation , as found
from the acceleration of a test mass towards a source mass, contains the term
1-G C( 2 ( 1 - 3I/mr 2 )(w · n ) 2 - 1-G [a 1 - C( 3 - a 2 ( 1 - J/mr 2 )] w 2 ,
where n is the unit vector along the line joining the two bodies and m, r and I
are the rest mass, radius and moment of inertia of the source mass, supposed
spherical .
If the sou rce density is uniform, I = �mr 2 and the first term is
- -fo-G a 2 (w n ) 2
·
(w is in units of the velocity of light) , while the second is

- 1-G(a 1 - a 3 -�cx 2 )w 2 .
Suppose an experiment is performed in a laboratory on the Earth's
surface, or in an artificial satellite, having a velocity v relative to the centre of
the Earth , itself having velocity w 0 in the preferred frame. Then
w w 0 + v,
=
w 2 w6 + v 2 + 2w 0 · v
=
and
(w n ) 2 [(w 0 + v) n] 2
· = ·
(w 0 n ) 2 + (v n ) 2 + 2(w 0 n)(v n ).
= •
·
•
·
If n is kept fixed in the laboratory frame, v n will be constant as will w 2 and

v 2 , but w 0 n and w 0 v vary with v and n .
·
· ·
This indicates that there might be two ways of doing experiments on

preferred frame effects - one would be to observe changes in G as the
laboratory goes round with the Earth or space craft, the other would be to
vary n periodically by turning the experiment in the laboratory.
As yet , no such experiments have been performed . If the difficulties of
setting up an experiment in a space craft could be overcome, that would be
the most promising scheme, for if the space craft were close to the Earth, the
direction of v would change with a period of about 100 minutes (50 minutes
if the experiment were symmetrical with respect to the masses) instead of
24 h for a terrestrial experiment, and the environment would probably be
much quieter.
66 A. H . Cook
The alternative, for a terrestrial experiment, is to change n . Chen (private

communication) has designed an experiment in which the position of a
source mass attracting the test mass of a torsion pendulum is moved
periodically, so changing n ; v is the velocity of the laboratory about the
centre of the Earth, so w 0 · v has a daily period, while v · n has the period with
which n is changed and w 0 n has periods equal to the sum and difference of
·
the two other periods, all thus readily identified in the observations.
No such experiment has yet been performed and the utility of them would
depend on whether they could attain appreciably better limits on the PPN
parameters than are set by observations on the Earth tides. As the Earth
rotates, the potential at any point on the surface varies because of the
changing positions of the Sun and Moon relative to the point, and because a
point on the surface does not coincide with the centre of mass of the Earth,
where the gravitational attraction of the Sun (or Moon) balances the inertial
acceleration in the orbit. The changing potential at the surface is the tidal
potential and can be calculated very accurately on Newtonian principles.
The tidal potential is about 1 part in 10 7 of the Earth's own attraction , and
gives rise to a variation in the magnitude of the local gravitational
acceleration of about 1 part in 10 7 , again accurately calculable. It is,
however, not the only effect. The Earth, being elastic, yields under the tidal
stresses, the direct consequence of which is that the radial distance of the
surface from the centre of mass changes, and hence also the gravitational
acceleration due to the mass of the Earth. The indirect consequence is that
the moment of inertia (and with it higher multi pole moments) changes with the
tidal potential , giving rise to a further change in gravity at the surface. The
elastic yielding of the Earth was at one time somewhat uncertain , but the
present knowledge of the distribution of density and elastic moduli within
the Earth (see Bullen and Bolt, 1985) is now such that the gravitational
effects of tidal yielding can be calculated quite well . A third effect is less
certain. The largest motions produced by the tidal potential are of course the
tides in the oceans, and they in turn have a direct and indirect effect upon the
surface gravity. The gravitational attraction at any point includes a
contribution from the oceans, and that changes with the tides. In addition ,
the pressure exerted by the water upon the surface of the solid Earth has a
tidal variation which leads to an additional distortion of the solid Earth, and
so to a further change in surface gravity. Finally, the tidal displacement of
the sea bed gives rise to yet further components of the ocean tides. All these
effects are again of the order of 1 part in 10 7 of surface gravity , but the ocean
tides and their effects cannot be calculated so well as those of the solid Earth
assuming the oceans absent. The ocean basins have complicated shapes, and
elaborate numerical integrations of the equations of motion are required for
meaningful results. It has also , until recently, been difficult to check the
solutions. Tides around the ocean basins are naturally well known, but
those far from land , which contribute most to the various gravitational
effects, could not be observed until fairly recently.
If there were a preferred frame of reference or a preferred location , the
local surface acceleration due to the gravitational attraction of the Earth as a
whole would vary as the Earth rotated and altered the alignment of the
direction of the centre of the Earth , and the position of the surface point
relative to the preferred position. Will ( 1 98 1) gives a thorough analysis of the
respective changes in gravity. The variations would mimic part of the Earth
tides and because the Earth tides are about 1 part in 10 7 of gravity, it can be
said that any preferred frame or location effects must be less than 1 part in
10 7 , and in fact less by at least one order of magnitude since to within an
order of magnitude the observed tidal effects agree with the calculations on
the best geophysical models.
The measurement of variations of 1 part in 10 8 in local gravity is
nowadays not so difficult, and the measurements themselves are not the
limitation in setting bounds on the PPN parameters. Spring gravity meters
as are commonly used in geophysical surveys are sufficiently sensitive, but
need to be provided with automatic recorders and isolated from mechanical
and thermal disturbance. With such precautions the drift of the gravity
meter can be kept low, as is necessary .for good tidal observations. Some
special instruments have been constructed for tidal observations, in
particular one in which a ball with a superconducting coating is suspended
electromagnetically in the field of superconducting magnet. The sensitivity
( 10 - 1 1 g) and stability are much superior to spring instruments. Analysing a
run of measurements at Pinon Flat in Southern California, Warburton and
Goodkind ( 1976) were able to set a limit of 10- 3 on rt 2 •
4.6 Additional deviations from general relativity

Besides the main effects discussed in the previous sections, there are others
that have been the subject of experiment or proposed experiment. First, it
was seen above that the general expression for the gravitational mass when a
body is considered as a source (active) is different from that when it is
considered as a test mass (passive). An experiment by Kreuzer ( 1 968) sets a
limit to such a difference. A cylinder of polytetrafluorethylene (Teflon) of
which fluorine comprises about 75 per cent by weight was placed in a tank
68 A. H . Cook
containing a mixture of trichlorethylene and dibromomethane adjusted in

composition (74 per cent of bromine by weight) so that it had the same
density as the Teflon cylinder which floated in it with neutral buoyancy.
Consequently, the passive gravitational masses in the gravity field of the
Earth of the Teflon and the liquid it displaced, were identical .
The Teflon cylinder was moved from side to side and the torsion balance
was used to detect any change of attraction on a test mass. At the limits of
the experimental sensitivity, it was found that the relative difference of the
ratio, active to passive gravitational mass, for fluorine and bromine was less
than 10 - 5 . Will ( 198 1) shows that the result entails for the PPN
parameter ( 3 ,
A possible dependence of gravitational force on temperature was

examined by Poynting and Phillips ( 1 905). A mass was heated to the
temperature of steam and cooled to that of liquid air and its weight (that is,
its gravitational attraction to the Earth) was found to change by less than 1
part in 10 1 0 per degree Centigrade, whether it was heated or cooled.
Gravitational shielding or non-linear superposition has also been
investigated. In Newtonian physics and general relativity there is no
shielding or polarising effect, and so , when a third body is interposed
between two others, it should not change the attraction between them.
It might also be that the gravitational attraction between two bodies
would depend on preferred directions (such as an optic axis) in the
crystalline structure. Mackenzie ( 1 895) used a torsion balance in direct
observations of attractions between anisotropic and isotropic bodies, but
detected no difference ; he also confirmed that, to within his accuracy, the
attraction was proportional to the product of the attracting masses.
Poynting and Gray ( 1 899) performed a further experiment in which they
attempted to excite rotation of a cylinder of crystalline quartz by rotating a
second cylinder close to it. They concluded that any directional effect
dependent on the crystal structure was less than 6 x 10 - 5 at 5.9 cm.
A laboratory experiment has been performed by Austin and Thwring
( 1897) who employed a torsion balance of similar design to that of Boys
( 1895) and found that when screens of lead, zinc, mercury, water or alcohol
were introduced , any reduction of the attraction was less than 1 part in 500.
Again , when the Moon comes between the Sun and the Earth at a total
eclipse of the Sun, the combined attraction of the Sun and the Moon at a
point on the Earth should be just the sum of the two separately. A total
eclipse has been seen near the sunset to the west of Trieste where the
variation of the horizontal attraction could be measured with very sensitive
horizontal pendulums of long period (Marussi in Cook , 19� 1). Nothing
inconsistent with the straight sum of the attractions before and after eclipse
was seen during eclipse. The test is not perhaps very rigorous for with a
similar geometrical disposition, electrostatic or magnetostatic screening
would be slight ; for substantial screening, the M oon would need to subtend
more nearly 90°.
A proposed test of the precession of a gyroscope in a space craft may be
considered as an experiment, rather than as an observation, for the
apparatus is under an experimenter's control . The experiment was first
suggested by Schiff ( 1960a, b) and has been in preparation for more than a
decade, but still awaits a suitable space flight. The rate of change of the spin
vector S is (Will , 198 1)
dS/dr = !l /\ S
where n is an angular velocity :
!! = - !v /\ a -!V A g + (y + !)v /\ V U
( U i s the gravitational potential , g i s a vector related to the metric tensor :
g = g0i ei (where ei i s a unit vector i n the }-direction) and a i s the spatial
component of the 4-acceleration and is zero in free fall , as in a satellite in
orbit about the Earth).
Thus if V is the centre of mass 3-velocity,
!l = i(4 + 4y + a 1 ) V /\ V -!a 1 w /\ V U + (y + !)v /\ V U ,
where w i s the velocity of the coordinate system relative to a preferred frame.
The observed precession is the resultant of a number of terms in the
equation for dS/dr . First , there is the geodetic precession, resulting from the
curvature of space near massive bodies. For a gyroscope in a circular orbit of
radius a about the Earth (of mass mE) the change of the direction of the spin
axis in one orbit is
JS = - 2n(y + 1-)(mE/a)(S /\ h),
where h is the unit vector perpendicular to the plane of the orbit. The rate is
approximately 7 arc seconds per year for a close satellite; and is much the
largest contribution to the precession.
The Lense-Thirring precession would be of the order of 0.0 1 arc seconds
per year, while a periodic precession arising from the velocity of a preferred
frame would have an amplitude of some 10 - 3 a 1 seconds.
The p ractical difficulties are very great. The prime requirement is that the
70 A . H . Cook
gyroscope should be subject to no torques other than those arising from the
effects sought, and that means that it must be uniform in shape and
homogeneous in density to about 1 part in a million. Secondly, the
recognition of the spin axis must involve no significant anisotropy or torque
and , lastly, the direction of spin has to be referred to a reference frame
constituted by distant stars, the directions of which should be established to
10 - 3 arc seconds. All these requirements present severe problems (Lippa
and Everitt , 1978 ; Cabrera and van Kann, 1978).
Lastly, in this section, the possible dependence of the constant of
gravitation upon time is mentioned, if only to remark that at present it seems
most unlikely that any dependence could be examined experimentally.
General relativity treats G as a fundamental constant, part of the system of
definition of the theory, and so not susceptible to experimental test (the
velocity of light in special relativity occupies a similar position) but other
theories admit variation, with the implication that G is not an independent
constant of nature, but a quantity related in some way to other constants of
nature so that it may meaningfully be compared with them to establish
whether or not it changes. Dirac's hypothesis of large numbers leads to a
rate of change of an order of magnitude in the lifetime of the universe, an
amount that · could only be checked by reference to geophysical or
astronomical observations. Analyses of, for example, p alaeomagnetic data
about possible changes in the radius of the Earth have been inconclusive.
Within the laboratory, an experiment on the time dependence of G would
need to establish the value of G in relation to the standards of length, time
and mass (in actual current practice, the velocity of light , the standard of
frequency and the standard kilogram) with sufficient precision at different
times to enable a change of 10 - 1 0 per year to be detected . Since, as will be
seen , the value of G is known at best to rather better than 1 part in 104, and
since the standard kilogram is established to no better than 1 part in 109 , it
will be clear that there is no prospect of any realistic experimental check
upon a possible variation of G in time.
4. 7 The measurement of the constant of gravitation

The measurement of the constant of gravitation has a curious place in
physics. On the one hand, the value of G is relatively poorly known
compared with other natural constants, such as Planck's constant , or the
mass of the electron , most of which are known to 1 part in 106 or better. The
reasons for this state of affairs have already been alluded to : gravitational
forces are very small compared with the electromagnetic forces and so
experiments are very susceptible to extraneous disturbances, while the
reliable and sensitive electromagnetic techniques that can be applied to
other constants are not available for gravitation. Yet, while the measured
value of G is so uncertain the relevance of G to the rest of physics is slight.
The other principal constants of physics form an interconnected set and a
good knowledge of their values has consequences in both fundamental
theory (relativistic quantum mechanics, for example) and in practical
measurement of high precision (establishment of standards for electrical
measurements as an instance). Almost no such requirements or implications
apply to knowledge of the value of G. It is, so far as is known or postulated
(but see the previous section) independent of all the other constants and is
not required in any practical system of measurement . It enters into the
conversion of orbital parameters in celestial mechanics into masses (in
kilograms) of celestial bodies, but no theory of the constitution of such
bodies is so well established in detail as to require the masses to be known in
terms of the terrestrial standard to better than a few per cent. The only
possible exception to that statement has to do with the internal constitution
of the Earth. The variations of density and elastic moduli within the Earth
are nowadays established to about 1 part in a thousand throughout most
of the Earth from analysis of numerous seismological observations. In essence
those analyses give the value of the ratios K/p and µ/p , where K is the bulk
modulus, µ the shear modulus and p the density. In those ratios the standard
of mass enters both the numerators and denominators and so appears in
neither of the ratios. At the same time, the absolute value of density is set by
the condition that the density must integrate up to the total mass of the
Earth, which is fixed by the attraction of gravity of the Earth ; the conversion
of density to units of the terrestrial standards of mass and length therefore
requires a knowledge of the constant of gravitation. The point of such a
conversion is to make comparisons with laboratory determinations of
equations of state of possible constituents of the body of the Earth, in which
the density is measured against the usual terrestrial standards and is
independent of the value of G.
1
Five methods have been used for the laboratory determination of the
constant of gravitation. The earliest uses the deflexion of a torsion balance
to measure the force between two masses. In the second method , a torsion
balance is used as an oscillator, the gravitational potential of attracting
masses being used to alter the restoring force and hence the period of
oscillation. In a variant of those two methods, the attracting masses were
72 A. H. Cook
moved back and forth around the torsional pendulum in resonance with the
period of oscillation of the pendulum and the steady amplitude of the
oscillations was measured. The force exerted by an attracting mass on a test
mass can also be measured by balancing it against the gravitational
attraction of the Earth (with a known acceleration due to gravity) on a
sensitive equal arm chemical balance. Lastly, a method first suggested in
principle by Newton has been attempted . Newton (see Section 4. 1)
calculated the time it would take two masses moving freely to come together
from an initial separation ; the experiment of Luther et al. ( 1976) is
somewhat different, in that the separation of two masses was kept constant
by a servo-system which moved the one away. from the other in opposition
to the gravitational acceleration. One mass (the controlled one) was on a
rotating table, the other was suspended from a torsion balance and the
servo-control of the first mass was adjusted to maintain deflexion of the
torsion balance constant , when the acceleration of the masses towards each
other was equal to the acceleration of the controlled mass.
In the eighteenth century, estimates of the constant of gravitation were
also made from the gravitational attraction of some isolated mountain (in
Scotland, Schiehallion and Arthur's Seat) which was found from the amount
by which a plumb bob was deflected towards the mountain - the tangent of
that angle is the ratio of the attraction of the mountain to that of the whole
Earth. The radical weakness of the method is the determination of the
attraction of the mountain , for the internal distribution of density can never
be known with any exactness and there is no such thing as an i solated
mountain with a clearly defined boundary. The interest of such observations
lies in what they tell us about sub-surface distributions of mass associated
with mountains and not in the values of G that have been derived .
All experimental determinations of the constant of gravitation suffer from
the same problems - the measurement of the very small gravitational forces
in the face of noise from other causes, and the location of the centres of mass
of the attracting and test masses and the measurement of their separations.
In the torsion balance, the force on the test mass is balanced by the torque
corresponding to the twist of the fibre of the balance, and that is so whether a
steady value of the twist is measured , as in the original work of Cavendish,
or whether the torque is the restoring force in free oscillations of the
pendulum , as in the determinations of Braun and others following him . The
essential advantage of finding the torque from the period of free oscillations
is that, if the pendulum is allowed to oscillate for many periods, a period can
be found much more precisely than a steady deflexion. In either case, the
torsional constant of the pendulum has to be found, usually from the free
period in the absence of attracting masses. The other problem common to all
experimental determinations is that the distribution of density within
masses may not be uniform and therefore the position of the centre of
attraction cannot be found from the dimensions of the exterior surface. That
problem can to a large extent be overcome by tu rning the attracting masses
so that the centre of attraction rotates about some known axis, and taking
observations in different positions. The test mass is usually made small
enough that the density can be assumed to be uniform. There is also the
problem of measuring from a fixed attracting mass to a mobile test mass, but
it usually is possible to arrange the observations so that the critical distances
are those between different positions of the attracting masses (compare the
experiment of Chen on the inverse square law - Section 4.3).
In all measurements with the torsion balance, the stability of the position
of rest of the balance beam is criticaJ. The measurement of a period is much
less subject to instability of the rest position, though not entirely so . The
cause of the instability is the movement of dislocations of the material of the
fibre under stress and a considerable improvement in technique was made
when Boys developed fused silica fibres which not only can be made much
finer than metal or other fibres, and so attain a greater sensitivity, but also
lack the dislocations of polycrystalline metals although as glasses they
undergo viscou s drift. Thu s, since the experiments of Boys, fibres of fused
silica have generally been used for the most delicate balances, although in
recent years B raginski and Manukin ( 1977), followed by Chen (Chen, Cook
and Metherell , 1984), have shown how to remove dislocations from
tungsten wires by a combination of thermal and mechanical stress, and so
have produced very stable torsion balances.
All the earlier determinations of the constant of gravitation depended on
the deflexion of a torsion balance (Cavendish , 1798 ; Baily, 1842 ; Cornu and
Baille, 1873, 1 878) and perforce were performed in an enclosure with air at
atmospheric pressu re ; they were greatly disturbed by convection cu rrents.
Two important advances were made by Boys ( 1 895) and Braun ( 1 897). Boys
introduced quartz fibres for the torsion suspension and showed that it was
an advantage to reduce the size of a deflexion experiment. Braun, who
worked in relative isolation in his Jesuit house, was the first to evacuate the
enclosure in which the torsion balance hung and also performed the first
experiment using the period of the balance instead of its deflexion - he used
the deflexion method as well.
Braun's use of the period of the torsion balance has been followed by Heyl
74 A. H . Cook
( 1930), Heyl and Chrzanowski ( 1942) and Luther and Towler ( 1982). If
attracting masses are placed in line with the beam of the torsion balance, the
torque they exert is in the same sense as that of the suspension and increases
the frequency of oscillation, whereas if they are placed at right angles to the
beam, the attraction decreases the frequency. The method seems to have
given the best results so far, by far the most precise result being, it seems, that
of Luther and Towler (Table 4. 1).
A variant of the method of periods was introduced by Zahradnicek (1933).
Two coaxial torsion balances were set up, a light one with masses of about
9 g and a heavy one with masses of about 1 1 kg. The periods were nearly but
not quite the same. The heavy balance was set into free oscillations driving
the light balance through the gravitational attraction ; the value of G is found
from the amplitude of the forced oscillation of the light balance.
A very sensitive equal arm common balance was used by von Jolly ( 188 1)
and by Poynting ( 189 1). Von Jolly plact!d a container which could be filled
with mercury below one pan , while in Poynting's determination a large
block of lead was placed below the mass on one pan of the balance and the
increase in weight of the mass was found by simple weighing. Poynting
devised a very sensitive means of measuring the deflexion of the balance and
hence the small increase in weight.
The accuracy of his result is comparable with that of other determinations
of that time (Table 4. 1). It is clear that if the sensitivity of the balance were to
Table 4. 1 . Determinations of the constant of gravitation
Result
Author Method ( 10 - 1 1 Nm 2/kg 2)
Cavendish, 1798 Torsion balance, deflexion 6.754

Poynting , 189 1 Common balance 6.698
Boys, 1895 Torsion balance, deflexion 6.658
Braun, 1 897 Torsion balance, period and
deflexion 6 .658
Heyl , 1930 Torsion balance , period :
gold 6.678
platinum 6.664
glass 6.674
Zah radnicek, 1933 Torsion balance, resonance 6.659
Heyl and Chrzanowski, 1942 Torsion balance, period :
hard drawn wire 6.676
annealed wire 6.668
Luther and Towler, 1982 Torsion balance, period 6.6726
be improved, the measurement of the distance between the centres of mass of

the weight on the pan and the heavy mass placed below it would be a
limitation. Deflexions of the apparatus as a result of putting the heavy
attracting mass in position and taking it away may also be a severe problem.
Lastly, Luther et al. ( 1976) developed the method first thought of by
Newton to measu re directly the acceleration of two nearby masses. Whereas
Newton estimated the time it would take for two masses to come together,
Luther and his colleagues devised a servo-system that kept the two masses
always at the same distance from each other, and measured the acceleration
of two nearby masses. A final result seems not to have been published.
4.8 Conclusion
The experiments discussed in this article are those that have been made in a
laboratory on Earth or that could be made in one in a space craft. Thus
nothing has been said of attempts to observe gravitational radiation , for,
although they use some of the most advanced techniques of experimental
physics, they depend upon radiation from sources which are clearly not
under the control of the experimenter; the experiments considered here
involve for the most part sources within the laboratory and at the
experimenter's disposition . In addition the techniques are all similar,
involving delicate mechanical measurements of very small forces. Tests of
general relativity or other theories of gravitation by means of analyses of
celestial mechanics and geophysics are also not considered.
The limitation to experimental tests means that the range of distances
over which theories can be checked 'is of the order of 1 m or somewhat less.
Taken with geophysical and astronomical observations (for the most part in
the solar system), laboratory experiments test the validity of theories of
gravitation from a few centimetres out to the extent of the solar system, so
that there are ranges of both larger and small distances for which there is no
experimental test.
Newton's law of gravitation implies that the attraction between two
bodies is proportional to mass and independent of material constitution,
that the law is the inverse square law and that G is a constant independent of
time, or any preferred coordinate ' frame or any preferred position.
Experimental results confirm the first implication at least to the extent that
the ratio of passive gravitational mass to inertial mas� is the same for all
materials to within 1 part in 10 1 2 , and they confirm that the law is the inverse
square to within 1 part in 10 4 . Geophysical evidence from Earth tides
indicates that any effects of a preferred frame or location are small and
76 A . H . Cook
nothing useful can be said about variation in time. There is some evidence
against screening of a gravitational source by other material.
Looked at from the point of view of PPN theory, observations on the
scale of the solar system show that the parameters y and /3 must have values
close to unity as predicted by general relativity, but the only other
conclusions to be drawn from experiment or observation is that the
parameter e< 2 is zero and that certain combinations of other parameters are
also zero. All parameters other than f3 and y are zero in general relativity.
In a number of instances, the results of experiments show unexplained
systematic effects, of which three have already been mentioned . The results
of the experiment by Roll, Krotkov and Dicke ( 1964) showed a 12 h period
in the recorded signal ; that cannot arise from the attraction of the Sun on the
masses and its origin has not been accounted for. Heyl ( 1930) , in his
determination of the gravitational constant, used different materials for the
masses in various groups of observations and found different results for G
(Table 4. 1), which is inconsistent with the Eotvos group of experiments.
Similarly, there appear to be some systematic differences between materials
within the Eotvos experiment itself, even though the interpretation
proposed by Fishbach et al. ( 1986) may be invalid. Again , Heyl and
Chrzanowski ( 1.942) found that whether the tungsten torsion fibre was hard
drawn or annealed gave different values of G (Table 4. 1). The significance of
these examples is not so much the physical interpretation that may be put
upon them, rather it is that it is difficult to attain an adequate understanding
of experiments at the limit of available techniques.
In view of the rather limited nature of experiments so far performed,
and in view of the difficulties in understanding the outcomes of
these experiments, there must be considerable interest in devising
new experiments, both better ones to reexamine existing results and new
ones to investigate other aspects of gravitation . Braginski et al. ( 1977)
have proposed three new techniques of detection which they argue should
be very sensitive and have suggested a number of experiments of which they
might form the basis. None of the experiments has, however, yet been
performed.
The naive experimenter may be allowed to conclude that general relativity
remains the best description of gravitation , but he is also well aware that
much ingeµuity, care and imagination will be required before his
experiments have the delicacy that will enable them to contribute a s
' '
Poynting hoped , to the foundations of the explanation of gravitation .

References
Anderson, J . D . , Esposito , P. B . , M artin, W . , Thornton , C. L. and Muhleman , D. 0 .

( 1975). Experimental test o f general relativity using time-delay data from Mariner 6
and Mariner 7 . Astrophys. J., 200, 22 1-3 3 .
Austin , L . W . and Thwring, C . B . ( 1897). A n experimental research o n gravitational
permeability. Phys. Rev., 5, 294-300.
Baily, F. ( 1 842). An account of some experiments with the torsion rod for determining
the mean density of the Earth. Mon. Not. R. Astron. Soc. , 5, 197-206.
Bessel , F. W. ( 1832). Versuche iiber die Kraft mit welcher die Erde Karper von
verschiedene Beschaffenheit anzieht. Ann. Phys. Chem. (Poggendorf), 25, 40 1- 17.
Boys, C . V . ( 1895). On the Newtonian Constant of Gravitation. Phil. Trans. Roy. Soc.,
A 182, 1-72 .
Braginski, V . B . , Caves, C. M . and Thorne, Kip. S. ( 1977). Laboratory experiments to
test relativistic gravity . Phys. Rev., D I S , 2047-68 .
Braginski, V . B . and M anukin, A. B. ( 1977). In Measurements of Weak Forces in Physics
Experiments, ed . D . H . Douglass. Chicago University P ress : Chicago.
Braginski, V. B. and Panov, V. I . ( 1972) . Verification of the equivalence of inertial and
gravitational mass. Sov. Phys. J.E . T.P., 34, 463--6.
Braun , C. ( 1897). Die Gravitationskonstante, die Masse und M ittlere Dichte der Erde .
Denschr. Akad. Wiss. (Wien), Math. naturwiss. Kl., 64, 187-258.
Bullen, K . E. and Bolt, B . A . ( 1985) . An Introduction to the Theory of Seismology.
Cambridge University Press : Cambridge.
Cabrera, B. and van Kann , F. J. ( 1978). Ultra-low magnetic field apparatus for a
cryogenic gyroscope. Acta Astronautica, 5, 125-30.
Cavendish , H . ( 1798). Experiments to determine the density of the Earth. Phil. Trans.
Roy. Soc. , 88, 469-526.
Chan, H . A . , Moody, M . V. and Paik, H . J . ( 1982). Null test of the gravitational
inverse square law . Phys. Rev. Lett., 49, 1745-8 .
Chen, Y . T. ( 1982). The gravitational field inside a long hollow cylinder of infinite
length . Proc. Roy. Soc. , A382, 75-82 .
Chen , Y . T., Cook, A . H . and Metherell , A . J . F. ( 1 984) . An experimental test of inverse
square law of gravitation. Proc. Roy. Soc., A394, 47-68.
Cook, A . H . ( 197 1). The experimental determination of the constant of gravitation. Nat.
Bur. Stds. Sp. Publ., 343, 475-83 .
Cook, A . H . and Chen, Y . T. ( 1982) . O n the significance o f the radial gravitational force
of the finite cylinder. J. Phys., AlS, 1 5 9 1-7 .
Comu , A . and Baille, J . ( 1873). Determination nouvelle de la constante de !'attraction et
de la densite moyenne de la Terre . C.R . Acad. Sci. (Paris), 76, 954-5.
Cornu, A. and Baille, J. ( 1 878). Sur la mesure de la densite moyenne de la Terre. C.R.
Acad. Sci. (Paris), 86, 699-702.
Eotvos, R . V . , Pekar, V . and Fekete, E. ( 1922). Beitrag zum Gesetze der
Proportionalitat von Tragheit u nd Gravitat . Ann. Phys., 68, 1 1-66.
Fishbach, E . , Sadarsk, E., Szafer, A . , Talmadge, C. and Aronson , S. H . ( 1986).
Reanalysis of the Eotvos experiment. Phys. Rev. Lett., 56, 3-6 .
Formalont, E. B . and Sramak, R . A . ( 1977). The deflection of radio waves by the Sun.
Comm. Astrophys., 7, 19-33.
78 A. H . Cook
Fujii, Y. ( 197 1). Dilation and possible non-Newtonian gravity. Nature Phys. Sci. , 234,
5-7 .
Fujii, Y. ( 1972). Scale invariance and gravity of hadrons. Ann. Phys., 69, 494-52 1 .
G ibbons, G. W. and Whiting, B. F . ( 1 98 1). Newtonian gravity measurements impose
constraints on unification theories. Nature (London), 291 , 636-8 .
Heyl, P. R . ( 1930). A redetermination of the constant of gravitation. J. Res. Nat. Bur.
Stds., 5, 1243-90.
Heyl, P . R . and Chrzanowski, P . ( 1942). A new determination of the constant of
gravitation. J. Res. Nat. Bur. Stds., 29, 1-3 1 .
Holding, S . C. and Tuck, G . J . ( 1984). A new mine determination o f the Newtonian
gravitational constant. Nature (London), 307, 7 14-16.
Keiser, G. M. and Faller, J . E. ( 1979). A new approach to the Eotvos experiment. Bull.
Amer. Phys. Soc., 24, 579.
Keyser, P. T., Niebauer, T. and Faller, J. E. ( 1 986). Comment on 'Reanalysis of the
EOtvos experimen t'. Phys. Rev. Lett., 56, 2425.
Kreuzer, L . B. ( 1968). Experimental measurement o f t h e equivalence o f active and
passive gravitational mass. Phys. Rev., 169, 1 007-12.
Lippa, J . A . and Everitt, C. W. F. ( 1978). The role of cryogenics in the gyroscope
experiment. Acta Astron., 5, 1 1 9-123.
Long, D. R . ( 1976). Experimental examination of the gravitational inverse square law.
Nature (London), 260, 4 1 7-18.
Long, D. R . ( 1 980). Nuovo Cim., 855, 252.
Luther, G. G. and Towler, W. R. ( 1982). Redetermination of the Newtonian
gravitational constant, G. Phys. Rev. Lett . , 48, 12 1-3 .
Luther, G. G . , Towler, W. R . , Deslattes, R . D . , Lowry, R . and Beams, J . ( 1976). Int.
Conf on Atomic Masses and Fundamental Constants 5, ed . J : H. Sanders and A. H .
-
Wapstra, p . 592. Plenum : New York.

Mackenzie, A . S. ( 1 895). On the attraction of crystalline and isotropic masses at small
distances. Phys. Rev . , 2, 32 1-43 .
Maskelyne, N. ( 1 775). A proposal for measuring the attractions of some hill in the
kingdom by astronomical observations. Phil. Trans. Roy. Soc., 65, 495-9.
Newton , I . ( 1 687). Philosophiae Naturalis Principia Mathematica, III. P rop. VI, Th. V I .
Ogawa, Y . , Tsubona , K . and Hirakawa, H . ( 1 982). Experimental test o f the law of
gravitation. Phys. Rev. D26, 729-34.
Panov, V. I . and Fron.tov , V. N. ( 1979). The Cavendish experiment at large distances.
Soviet Physics, J.E.T.P., 50, 852-6.
Poynting, J. H. ( 1 89 1 ). On the determination of the mean density of the Earth and the
gravitation constant by means of the common balance. Phil. Trans. Roy. Soc., A l82,
565-656.
Poynting, J. H . ( 1900). Recent studies in gravitation . Proc. Roy. Inst. G.B., 16, 278-94.
Poynting, J. H . and Gray, P. L. ( 1 899). An experiment in search of a directive action of
one quartz crystal by another. Phil. Trans. Roy. Soc., Al92, 245-56.
Poynting, J. H . and Phillips, P. ( 1905). An experiment with the balance to find if change
of temperature has any effect on weight. Proc. Roy. Soc., A 76, 44 5- 5 7 .
Reasenberg, R . D . , Shapiro , I . I . , MacNeil, P . E. , Goldstein, R . B . , Breidenthal, J . C.,
Brenkle, J . P . , Cain, D . L., Kaufman , T. M . , Komarek , T. A . and Zygielbaum, A. I.
(1979). Viking relativity experiment : Verification of signal retardation by solar gravity.
Astrophys. J., 234, 2 1 9-2 1 .
Renner, J . ( 1935). Hung. Acad. Sci., 53 (II), 542-70.
Roll , P. G . , Krotkov, R. and Dicke, R . H . ( 1964). The equivalence of inertial and

passive gravitational mass. Ann. Phys . (New York), 26, 442-5 1 7 .
Schiff, L . I . ( 1960a). Motion o f a gyroscope according t o Einstein's general theory of
relativity . Proc. Nat. Acad. Sci. ( USA), 46, 87 1-82 .
Schiff, L . I . ( 1960b) . Possible new test of general relativity theory. Phys. Rev. Lett., 4,
2 1 5- 17.
Shapiro, I. I., Counselman, C. C. III and King, R. W . ( 1976). Verification of the
principle of equivalence for massive bodies. Phys. Rev. Lett., 36, 555-8.
Shapiro, I. I . , Pettengill, G. H., Ash, M . E., lngills, R. P., Campbell, D. B. and Dyce,
R. B. ( 1972). Mercury's perihelion advance - determination by radar. Phys. Rev. Lett. ,
28, 1 594-7.
Spero, R . E., Hoskins, J. K . , Newman, R., Pellam, J. and Schultz, J. ( 1980). Tests of the
gravitational inverse square law at laboratory distances. Phys. Rev. Lett., 44, 1645-8 .
Stacey , F. D . and Tuck, G . J. ( 198 1 ) . Geophysical evidence for non-Newtonian gravity.
Nature (London), 292, 230-2.
von Jolly , Ph. ( 188 1 ). Die Anwendung der Wage auf Probleme der Gravitation. Ann. der
Physik u. Chemie ( Wiedener), N .F., 14, 33 1-55.
Warburton, R . J. and Goodkind, J . M . ( 1 976 ) . Search for evidence for a preferred
reference frame. Astrophys . J., 208 , 88 1-6 .
Will , C . M . ( 1 98 1 ) . Theory and Experiment in Gravitational Physics . Cambridge
Yu, H-T., Ni, W-T . , Hu, C-G ., Liu , F-H . , Yang, C-H . and Lin , W-N . ( 1979).
Experimental determination of the gravitational forces at separations around
10 meters. Phys. Rev., 020, 1 8 1 3- 1 5 .
Zahradnicek , J . ( 1 933). Resonanz methode fiir die Messiing der Gravitations konstante
mittels der Drehwaage. Phys. Zeit., 34, 126-33 .
5
Experimental gravitation from Newton 's
Principia to Einstein's general relativity
CLIFFORD M .W I LL
5.1 Introduction
The tercentenary of the publication of Newton's Principia has occurred at a
propitious time in the subject of gravitation physics. We find ourselves in the
midst of a renaissance for general relativity, the theory of gravitation that
superseded Newtonian gravity, and that is now one of the most active and
exciting branches of physics. General relativity has become an important
theoretical tool for the astrophysicist, as well as a fundamental ingredient in
the quest for unification of the theories of the basic interactions.
Like any other branch of physics, gravitation has a strong experimental
component as well . The confrontation between theory and experiment was a
central element in the Principia ; there Newton reported results of his own
experiments to verify the principle of equivalence, and made detailed
comparisons of the predictions of his theory of gravity with astronomical
observations. In the two centuries following publication of the Principia,
Newtonian gravitation was put to the test in a variety of ways, and passed
every test but one with flying colors. Nor did Einstein shy away from
experimental confrontation, for he proposed three important tests of general
relativity. Although two of Einstein's three tests were confirmed
immediately or shortly after publication of his theory in 19 16, further
experimental progress during the next 45 years was very slow, largely
because of a lack of experimental technology of sufficient accuracy to
measure the extremely small predicted effects.
However, during the two decades 1960--80, there was a rebirth in the
subject of experimental gravitation . This rebirth coincided with the
renaissance of general relativity as a whole, and was propelled by
astronomical discoveries that indicated a role for relativistic gravity in
astrophysics, by new theoretical insights into the observable consequences
Experimental gravitation from Newton to Einstein 81
of general relativity, and by advances i n the technology of laboratory and

space experimentation that provided high-precision tools to carry out
gravity tests.
The period 1960--80 became the 'decades for testing general relativity'.
The systematic high-precision testing of gravitational theory emerged as an
active and challenging field , with improved versions of Einstein's tests such
as the deflection of light and the gravitational redshift, and brand new tests,
such as the Shapiro time delay and the Nordtvedt effect in lunar motion .
Several sophisticated theoretical frameworks were developed to compare
the relative merits of various experiments, to help discover new tests and to
classify and compare alternative theories of gravitation.
As did Newtonian theory in earlier years, so too did general relativity pass
virtually every experimental test with equally flying colors. Although the
entire story of experimental gravitation is far from over, it is reasonable to
say that the second chapter of the story has been written. The first chapter
consisted of the verification of Newtonian gravity, between 1687 and the end
of the nineteenth centu ry. The second chapter consisted of the verification of
the principal predictions of general relativity, with the peak of activity
during the decades 1960--80. The next chapter of the story puts us at the
frontiers of experimental gravitation , where the observable effects are very
small and the experiments very difficult. The effects to be detected
correspond in some sense to the finer predictions of general relativity, such
as the dragging of inertial frames and its effect on orbiting gyroscopes, or the
presence of higher-order effects in phenomena such as the deflection of light.
We may enter the next century before this chapter is concluded.
This is therefore an appropriate occasion for a summary and
consolidation of the experimental evidence that forms the empirical
foundation both for Newtonian gravitation as well as for general relativity.
We shall begin this summary by placing the two respective theories in the
proper context. Newton's and Einstein's conceptions of gravity had a
number of common elements , such as the equivalence principle, yet differed
radically in other elements, such as the nature of the spacetime underlying
gravity. Section 5.2 reviews the commonalities and contrasts between the
two theories. In Section 5.3, we discuss four fundamentals of gravitation
that, in some ways, are common to both theories : the principle of
equivalence, the inverse square law, the equality of action and reaction, and
the constancy of the gravitational constant, and review the experimental
evidence supporting them. The testable differences between the theories
show up in applications where general relativity mainly predicts corrections
82 C. M. Will
to the Newtonian behavior. In Section 5.4, we describe these applications,

specifically to the motions of planets, the motion of the M oon, the motion of
light, and the motion of binary star systems, and describe the extent to which
Newtonian gravity and general relativity have been verified . Section 5.5
presents concluding remarks. In an Appendix, we summarize the
parametrized post-Newtonian (PPN) formalism , which is a useful
framework for discussing experimental gravitation in the solar system .
Our goal is not to give an exhaustive, completely up-to-date technical
review of this subject with a complete bibliography, rather it is to honor
Newton on this occasion by demonstrating the remarkable success of
gravitational theory, both the Newtonian and the Einsteinian , in the
confrontation with experiment. As a result , we will not provide complete
references to work done in this field, but instead will refer the reader to the
appropriate technical review articles and monographs, specifically to
Theory and Experiment in Gravitational Physics (Will, 198 1), hereafter
referred to as TEGP, and to 'The Confrontation between General Relativity
and Experiment : an Update' (Will, 1984), hereafter referred to as UPDATE.
Other references will be confined to the key original papers, and to
important recent papers that are not included in TEGP or UPDATE. A
popular account of this subject , with historical sidelights and anecdotes, can
be found in Was Einstein Right? (Will, 1986).
5.2 Newtonian gravitation and general relativity : commonalities and

contrasts
Newtonian gravitational theory and general relativity are radically different
theories of the gravitational interaction . Newtonian theory is founded upon
the notion of absolute space and absolute time, while general relativity is
based upon spacetime and the compatibility with special relativity . In
Newtonian physics, the permissible coordinate transformations are those of
the Galilei group , while in general relativity, arbitrary coordinate
transformations are allowed . The gravitational potential in Newtonian
theory is a field that is 'pasted on' the absolute space and time underlying the
theory, while in general relativity, the gravitational potential is the
spacetime metric itself, and the effects of gravity are consequences of curved
spacetime. Since the gravitational interaction in Newtonian theory is
assumed to be instantaneous (although Newton himself never made this
assumption) gravitational waves do not exist , but, more than this, the actual
causal structure of the theory is very different from that of general relativity,
whose causal structure is based upon null cones or light cones that reflect the
finite propagation speed of all interactions, including gravitation.

Newtonian gravitation is a linear theory, whereas general relativity is highly
non-linear. These differences are illustrated by the kinds of topics in general
relativity that are discussed in other articles in this volume : black holes,
gravitational radiation, cosmological models, geometry of spacetime, and
so on, for which the predictions of Newtonian theory are very different from
those of general rel a ti vi ty, or are even non-existent.
On the other hand, there are many ways in which the two theories are
similar. Both theories are based on a principle of equivalence whose main
element is the concept that bodies fall with the same acceleration regardless
of their structure or composition. Both theories embody the law of 'action
equals reaction', Newtonian theory by explicit assumption, general
relativity by being based on an invariant action principle, one of whose
consequences is the conservation of momentum or the equality of action and
reaction . In both theories, the fundamental constant that determines the
strength of gravity, the gravitational constant G, is a true constant of nature.
Even the inverse square law, one of the main tenets of Newtonian gravity,
can be seen to be a common feature of the two theories, if viewed properly.
To be sure , in general relativity, the inverse square law for the gravitational
force is valid only as a first approximation in situations of weak
gravitational fields, but there is a sense in which the inverse square law is an
exact parallel between the two theories. In Newtonian theory, the inverse
square law for the gravitational acceleration implies that the divergence of
the acceleration (V g), which is the same as the Laplacian of the potential
(V 2 U) , must vanish in vacuum. The same statement holds in general
·
relativity. There, in a local freely falling frame, the measured gradients of the
acceleration experienced by a collection of particles at rest are the 'electric'
components of the Riemann tensor ( R0rn) ; the divergence of the
acceleration is thus the trace of R o iOi • which is the Ricci tensor component
R00 , which vanishes in vacuum, according to Einstein's field equations.
Yet there is one area in which the two theories are very similar though
slightly different, and this is the area of prime importance for experimental
gravitation . This is the weak-field , slow-motion limit of general relativity,
the limit appropriate for discussing motions in the solar system and in other
astronomical arenas. In this limit, the Einstein field equations can be solved
by successive approximations. At the first approximation, the field
equations and equations of motion for matter are equivalent to those of
Newtonian gravity. This is the 'Newtonian limit' of general relativity. At the
next level of approximation, known as the 'post-Newtonian
84 C. M. Will
approximation', correction terms occur that lead to the important,

experimentally observable effects such as the perihelion advance of
Mercury, and the deflection of light. It may soon be possible to devise
experiments to look for effects that enter at the next, or 'post-post
Newtonian' level of approximation.
One feature of the post-Newtonian approximation that has proved to be
very important in the discussion of experimental gravitation is the fact that
most alternative spacetime metric theories of gravitation to general
relativity, such as the Brans-Dicke scalar-tensor theory, have post
Newtonian limits with the same general structure, the only difference from
one theory to the next being the values of a set of numerical coefficients that
appear in the post-Newtonian metric. This circumstance has made it
possible to develop a very useful formalism , called the 'parametrized post
Newtonian' formalism for treating experimental tests and for analysing
alternative theories of gravity. The main features of this framework are
summarized in an Appendix.
The formal link between general relativity and Newtonian gravity has
received renewed attention recently. The issue is, can Newtonian gravity, as
formulated for example in the geometric, coordinate-free language of
Cartan, be obtained formally from general relativity as a limit of a sequence
of spacetimes? The limiting equations of motion are not at issue, rather it is
the limiting global structure of the spacetime that is the question. In
particular, how can one go from the causal structure of general relativity
with null cones and gravitational radiation to that of Newtonian gravity
with neither null cones nor gravitational radiation , along a smooth sequence
of spacetimes? Understanding this issue may contribute to the rigorous
foundation of equations of motion and of formulae for gravitational
radiation in the weak-field slow-motion limit of general relativity (Winicour,
1983 ; Futamase and Schutz, 1985).
One important commonality between Newtonian gravity and general
relativity : both have been subjected to strenuous experimental testing, and it
is to these tests that we now turn .
5.3. Testing the fundamentals of gravitation theory

5 .3 . 1 The principle of equivalence
The principle of equivalence has historically played an important role in the
development of gravitation theory. Newton regarded this principle as such a
cornerstone of mechanics that he devoted the opening paragraph of the
Principia to it , stating in Definition I , 'this quantity that I mean hereafter

everywhere under the name of . . . mass . . . is known by the weight . . for it is
proportional to the weight, as I have found by experiments on pendulums,
very accurately made . . . '.t In 1907, Einstein used the principle as a basic
element of general relativity. We now regard the principle of equivalence as
the foundation , not of Newtonian gravity or of general relativity, but of the
broader idea that spacetime is curved.
One elementary equivalence principle is the kind Newton had in mind
when he stated that the property of a body called 'mass' is proportional to
the 'weight', and is known as the weak equivalence principle (WEP). An
alternative statement of W EP is that the trajectory of a freely falling body
(one not acted upon by such forces as electromagnetism and too small to be
affected by tidal gravitational forces) is independent of its internal structure
and composition . In the simplest case of dropping two different bodies in a
gravitational field, the weak equivalence principle states that the bodies fall
with the same acceleration.
A much more powerful and far-reaching equivalence principle is known
as the Einstein equivalence principle (EEP). It states (i) WEP is valid, (ii) the
outcome of any local non-gravitational experiment is independent of the
velocity of the freely falling reference frame in which it is performed, and (iii)
the outcome of any local non-gravitational experiment is independent of
where and when in the universe it is performed. The second piece of EEP is
called local Lorentz invariance (LLI), and the third piece is called local
position invariance (LPI). For example, a measurement of the electric force
between two charged bodies is a local non-gravitational experiment; a
measurement of the gravitational force between two bodies (Cavendish
experiment) is not.
M ore so than the weak principle, EEP is the heart and soul of
gravitational theory, for it is possible to argue convincingly that if EEP is
valid , then gravitation must be a 'curved spacetime' phenomenon, in other
words, the effects of gravity must be equivalent to the effects of living in a
curved spacetime. As a consequence of this argument, the only theories of
gravity that can embody EEP are those that satisfy the postulates of 'metric
theories of gravity', which are (i) spacetime is endowed with a symmetric
metric, (ii) the trajectories of freely falling bodies are geodesics of that metric,
and (iii) in local freely falling reference frames, the non-gravitational laws of
t This and other quotations from the Principia are taken from the so-called 'Cajori'
edition (Cajori, 1934).
86 C. M. Will
physics are those written in the language of special relativity. The argument
that leads to this conclusion simply notes that, if EEP is valid, then in local
freely falling frames, the laws governing experiments must be independent of
the velocity of the frame (LLI), with constant values for the various atomic
constants (in order to be independent oflocation). The only laws we know of
that fulfil this are those that are compatible with special relativity , such as
Maxwell's equations of electromagnetism. Furthermore, in local freely
falling frames, test bodies appear to be unaccelerated, in other words, they
move on straight lines, but such 'locally straight' lines simply correspond to
'geodesics' in a curved spacetime (TEGP, section 2.3).
General relativity is a metric theory of gravity , but then so are many
others , including the Brans-Dicke theory. Because Newtonian gravitation
is not compatible with LLI , it is not a metric theory (although it can be
expressed in geometrical , coordinate-free language) . So the notion of curved
spacetime is a very general and fundamental one, and therefore it is
important to test the various aspects of EEP thoroughly .
A direct test of WEP is the comparison of the acceleration of two
laboratory-sized bodies of different composition in an external gravitational
field. Such tests of WEP predate Newton, including Philiponos (5th or 6th
C.), Stevin ( 1 586) and Galileo (c. 1 590). If the principle were violated, then
the accelerations of different bodies would differ. The simplest way to
quantify such possible violations of WEP in a form suitable for comparison
with experiment is to suppose that for a body with inertial mass m1 , the
passive gravitational mass mp is no longer equal to m1 , so that in a
gravitational field g, the acceleration is given by m1 a = mpg. Now the inertial
mass of a typical laboratory body is made up of several types of mass
energy : rest energy, electromagnetic energy, weak-interaction energy, and
so on. If one of these forms of energy contributes to mp differently than it
does to m1 , a violation of WEP would result . One could then write
mp = m1 + L, 11AEA/c 2 , ( 1)
A
where EA is the internal energy of the body generated by interaction A , and
>JA is a dimensionless parameter that measures the strength of the violation
of WEP induced by that interaction, and c is the speed of light. A
measurement or limit on the fractional difference in acceleration between
two bodies then yields a quantity called the 'Eotvos ratio' given by
1J
=
2la 1 - a i l
la 1 + a 2 I
( _ )
= L 1'/ A
Et E1
m i c2 mi c 2 •
(2)
A
Thus, experimental limits on Yf place limits on the WEP-violation

parameters YfA .
Many high-precision Eotvos-type experiments have been performed,
from the pendulum experiments of Newton , Bessel and Potter, to the classic
torsion balance measurements of Eotvos, Dicke, Braginsky and their
collaborators. Newton , for example, suspended identical boxes of wood
from wires 1 1 feet long. He filled one box with wood and the other with an
equal weight of gold and set the pair of pendula in motion, observing that
they kept in step to high accu racy. Repetitions of the experiment using other
materials led to the conclusion that the accelerations of different materials
are the same to about one part in a thousand . In the modern torsion balance
experiments, two objects of different composition are connected by a rod
and suspended in a horizontal orientation by a fine wire. If the gravitational
acceleration of the bodies differs, there will be a torque induced on the
suspension wire, related to the angle between the wire and the direction of
the gravitational acceleration g. If the entire apparatu s is rotated about
some direction with angular velocity w, the torque will be modulated with
peri od 2n/w. In the experiments of Eotvos and his collaborators, the wire
and g were not quite parallel because of the centripetal acceleration on the
apparatus due to the Earth's rotation; the apparatus was rotated about the
direction of the wire . In the Princeton (Dicke et al.) and Moscow (Braginsky
et al.) experiments, g was that of the Sun, and the rotation of the Earth
provided the modulation of the torque at a period of 24 h. The modulated
torque was determined either by measuring the torsional motions of the rod
or by measuring the force required to counteract the torque and keep the rod
in place. The resulting upper limits on measurable torques yielded limits on
Yf given by IYfl < 1 x 10 - 1 1 (Princeton , 1964) and IYfl < 1 x 10 - 1 2 (Moscow,
1972), where the limits are l a formal standard deviations (see TEGP,
section 2 .4(a), for further discussion and references). The primary sources of
error in these experiments are seismic noise and coupling of the torsion
balance to gradients in the external gravitational field . Attempts to improve
these results have centered on different forms of suspension of the masses,
including magnetic levitation, flotation on liquids, and potential use of
orbiting platforms. The results of various experiments are summarized in
Table 5 . 1 .
In order to determine the limits placed on individual parameters Yf A by
these experiments, it is necessary to estimate the coefficients E A/mc 2 for
different interactions and different materials. For laboratory-sized bodies,
the dominant contribu tion comes from the nucleus. Table 5.2 summarizes
88 C. M. Will
these estimates and limits (based on the Moscow experiments) for various
interactions (see TEGP, section 2.4(a), for details).
Recently, there has been a renewal of interest in Eotvos's original version
of the experiment as a consequence of a reanalysis of the Eotvos data by
Fischbach et al. ( 1986a). One of the goals of that reanalysis was to search for
the effects of a hypothetical short-range ( � 100 m) force, known as the 'fifth'
Table 5. 1 . Tests of the weak equivalence principlea
Experi- Substances Limit

ment Name Method tested on 1111
Newton Newton Pendula Various 10 - 3

Bessel Bessel Pendula Various 5 x 10 - 5
Eotvos Eotvos , Pekar Torsion balance Various 5 x 10 -9
and Fekete
Potter Potter Pendula Vari ous 2 x 10 - 5
Renner Renner Torsion balance Vari ous 2 x 10 - 9
Princeton Roll , Krotkov Torsion balance Aluminum and 10 - 1 1
and Dicke gold
Moscow Braginsky and Torsion balance Aluminum and 10- 1 2
Panov platinum
Munich Koester Free fall Neutrons 3 x 10 - 4
Stanford Worden M agnetic suspension Niobium, Earth 10 - 4
Boulder Keiser and Flotation on water Copper, tungsten 4 x 10 - 1 1
Faller
° For references see TEGP , section 2.4(a).
Table 5 .2. Experimental limits on equivalence principle violations for

various interactionsa
Interaction Limit on 11A from M oscow Eotvos experiment
Strong 5 x 10- 1 0
Electromagnetic
Electrostatic 4 x 10 - 1 0
M agnetostatic 6 x 10 - 6
Hyperfine 2 x 10 - 7
Weak 10 - 2
G ravitational No limit
° For discussion and references, see TEG P , section 2.4(a) .

force, that could couple to hypercharge or baryon number. Such a force

would cause a difference in acceleration between bodies whose baryon
number to mass ratio differed, and would yield a measurable Eotvos ratio
given by
(3)
where µi is the mass of the ith body measured in atomic mass units, Bi is its
total baryon number, and 71 Y is a parameter that depends on the strength
and range of the putative force, and on the detailed distribution of local
matter within that range. Fischbach et al. claimed that Eotvos's data
showed a significant dependence of 11 on the baryon-number-per-unit-mass
difference, and they quoted a value 71 Y = (5.65 + 0.7 1) x 10 - 6 . This result,
they argued , was qualitatively in accord with measured deviations from the
inverse square law of gravity using gravimeter data from deep mines
(Holding et al. , 1986), and with anomalous energy dependences in the
fundamental parameters that characterize the behavior of the K 0 - K.0
mesons , effects that would also be consequences of such a fifth force. The
more precise Princeton and Moscow experiments would not be sensitive to
this effect because the relevant source of gravity in those cases was the Sun,
and the effect of a 100 m short-range force would therefore be negligible.
A number of authors subsequently took issue with the results of this
reanalysis. It was argued by some that the evidence for the behavior
indicated by (3) was much weaker than claimed by Fischbach et al. because
of a number of factors which were inadequately taken into account in the
reanalysis, including a sign error in the interpretation of one of Eotvos's
conventions, uncertainties in the isotopic composition of each element and
the chemical composition of each compound used in the experiment,
uncertainties in the individual masses of the substances used and of the
containers in which they were placed, and unknown systematic errors that
might have affected the outcomes of Eotvos's experiments, given that three
different methods were used by Eotvos and his colleagues. Reanalyses by
others of Eotvos's data, and of the data from a series of 1935 experiments
using the same apparatus by Renner failed to support the Fischbach et al.
value of 71 r. Other authors pointed out that, even if one accepts the
dependence of 11 on baryon number claimed by Fischbach et al. , it is virtually
impossible to infer anything about the nature of a short-range fifth force,
because its putative effect on Eotvos's experiments is extremely sensitive to
the details of the nearby mass distribution (such as the mass of the building
90 C. M. Will
next to Eotvos's Budapest laboratory), which are unknowable nearly a

century after the fact. t At present the situation is inconclusive. However, the
Eotvos reanalysis has stimulated extensive interest in new experiments to
test this idea, and several are in progress.
The second ingredient of EEP, LLI , can be said to be tested every time
that special relativity is confirmed in the laboratory. However, many such
experiments, especially in high-energy physics, are not 'clean' tests, because
in many cases it is unlikely that a violation of Lorentz invariance could be
distinguished from effects due to the complicated strong and weak
interactions.
However, there is one class of experiments that have been interpreted as
'clean' tests of local Lorentz invariance, and with ultra-high precision at
that. This is the class of 'mass anisotropy' experiments : the classic version is
the Hughes-Drever experiment, performed in the period 1959-60
independently by Hughes and collaborators at Yale University, and by
Drever at Glasgow University (TEGP, section 2.4(b)); improved versions
have been carried out recently by Prestage et al. ( 1985) and by Lamoreaux
et al. ( 1986) . A simple and useful way of interpreting these ex
periments is to suppose that the electromagnetic interactions suffer a
slight violation of Lorentz invariance, through a change in the speed oflight
relative to the limiting speed of test particles, in other words, clight #- c 0
(TEGP, section 2.6; see also Haugan ( 1986b)). Such a violation necessarily
selects a preferred universal rest-frame, presumably that of the cosmic
background radiation , through which we are moving at about 300 km s - 1 .
Such a Lorentz-non-invariant electromagnetic inte raction would cause
shifts in the energy levels of atoms and nuclei that depend on the orientation
of the quantization axis of the state relative to our universal velocity vector,
and on the quantum numbers of the state. The presence or absence of such
energy shifts can be examined by measuring the energy of one such state
relative to another state that is either unaffected or is affected differently by
the supposed violation. One way is to look for a change in the energy levels
of states that are ordinarily equally spaced, such as the four J = 3/2 ground
states of the lithium-7 nucleus in a magnetic field (Orever experiment);
another is to compare the levels of a complex nucleus with the atomic
t Most of the papers commenting on the Fischbach et al. analysis were contained in the
2 June 1986 issue of Physical Review Letters : they include Neufield ( 1986), Thieberger
( 1986), Nussinov ( 1986) , Thodberg ( 1986) , Keyser et al. ( 1986) , together with responses
by Fischbach et al. ( 1986b) . Also listed by the editors are the authors of similar papers
which were not published for reasons of space; see also De Rujula ( 1986).
hyperfine levels of a hydrogen-maser clock (Prestage et al. experiment), and

another is to monitor the Larmor precession of mercury-20 1 nuclei in a
constant magnetic field (Lamoreaux experiment). These experiments have
all yielded extremely accurate results, quoted as limits on the parameter
2 2 1] in Table 5.3.
[(cl ight ) /(c 0 )
-
The principle of LPI , the third part of EEP, can be tested by the
gravitational redshift experiment , the first experimental test of gravitation
proposed by Einstein. Despite the fact that Einstein regarded this as a
crucial test of general relativity, we now realize that it does not distinguish
between general relativity and any other metric theory of gravity, instead it
is a test only of EEP. A typical gravitational redshift experiment measures
the frequency or wavelength shift Z = d v/v = dA/J,. between two identical
-
frequency standards (clocks) placed at rest at different heights in a static

gravitational field . If the frequency of a given type of atomic clock is the same
when measured in a local , momentarily comoving freely falling frame
(Lorentz frame) independent of the location or velocity of that frame, then
the comparison of frequencies of two clocks at rest at different locations
boils down to a comparison of the velocities of two local Lorentz frames, one
at rest with respect to one clock at the moment of emission of its signal , the
other at rest with respect to the other clock at the moment of reception of the
signal . The frequency shift is then a consequence of the first-order Doppler
shift between the frames. The structure of the clock plays no role
whatsoever. The result is a shift
Z = d U/c 2 , (4)
where d U is the difference in the Newtonian gravitational potential between
the receiver and the emitter. If LPI is not valid , then it turns out that the shift
can be written
Z = ( l + C( ) l'.\ U/c 2 , (5)
Table 5.3. Mass anisotropy tests of local Lorentz invariancea
Experiment Substances tested
Hughes-Drever Lithium-7 5x w- 1 6
P restage et al. Beryllum-9 10 - 1 8
Lamoreaux et al. Mercury-20 1 10 - 20
° For discussion see TEGP, sections 2.4(b), 2.6(e), and Haugan ( 1986b).
92 C. M. Will
where the parameter a. may depend upon the nature of the clock whose shift
is being measured (see TEGP, section 2.4(b), for details).
Although there were several attempts following the publication of the
general theory of relativity to measure the gravitational redshift of spectral
.
lines from white dwarf stars, the results were inconclusive (see Bertotti et al.
( 1962) for a review). The first successful, high-precision redshift
measurement was the series of Pound-Rebka-Snider experiments of
1960-5, that measured the frequency shift of y-ray photons from iron-57 as
they ascended or descended the Jefferson Physical Laboratory tower at
Harvard University. The high accuracy achieved - 1 per cent - was obtained
by making use of the Mossbauer effect to produce a narrow resonance line
whose shift could be accurately determined . Other experiments since 1960
measured the shift of spectral lines in the Sun's gravitational field and the
change in rate of atomic clocks transported aloft on aircraft, rockets and
satellites. Table 5.4 summarizes the important redshift experiments that
have been performed since 1960.
The most recent experiments have taken advantage of the development of
frequency standards of ultra-high stability - parts in 101 5-10 1 6 over
averaging times of 10- 100 seconds and longer, such as hydrogen-maser
clocks, and superconducting-cavity stabilized oscillator (SCSO) clocks.
Table 5 .4. Gravitational redshift experimentsa
Limit
Experiment Name Method on ia l
Pou nd-Rebka-Snider Pound and Rebka, Fall of photons from 10 - 2

Pound and Snider Mossbauer emitters
Brault Brault Solar spectral lines 5 x 10 - 2
Jenkins Jen kins Crystal oscillato r 9 x 10 - 2
clocks on G EOS- 1
Satellite
Snider Snider Solar spectral lines 6 x 10 - 2
Jet-lagged clocks (A) Hafele and Cesium beam clocks 10 - 1
Keating on jet aircraft
Jet-lagged clocks (B) Alley Rubidium clocks on 2 x 10 - 2
jet aircraft
Vessot-Levine rocket Vessot et al. Hyd rogen maser on 2 x 10 - 4
redshift experiment rocket
Null redshift experiment Turneaure et al. Hydrogen maser vs. 2 x 10 - 2
scso
a For references see TEG P , section 2 .4(c) .

The first such experiment (and the most precise to date) was the Vessot
Levine rocket redshift experiment that took place in June 1976 (in NASA
circles, the experiment was denoted Gravity Probe A or GPA) . A hydrogen
maser clock was flown on a rocket to an altitude of about 10 000 km and its
frequency compared with a similar clock on the ground. The experiment
took advantage of the masers' frequency stability by monitoring the
frequency shift as a function of altitude. A sophisticated data acquisition
scheme accurately eliminated all effects of the first-order Doppler shift due
to the rocket's motion, while tracking data were used to determine the
payload's location and velocity (to evaluate the potential difference .1 U, and
the time dilation). Analysis of the data yielded a limit l al < 2 x 10 - 4 (V essot
et al. , 1980). Improvement of this limit may be possible by placing such
clocks on Earth-orbiting satellites or on a spacecraft in a very eccentric solar
orbit.
Advances in stable clocks have also made possible a redshift experiment
that is a more direct test of LPI : a 'null ' gravitational redshift experiment
that compares two different types of clocks, side by side in the same
laboratory . If LPI is violated , then not only is the proper ticking rate of an
atomic clock dependent upon position, but the position dependence must
itself depend on the structure and composition of the clock, otherwise all
clocks would vary with position in a universal way and there would be no
operational way to detect the effect (since one clock must be selected as a
standard and ratios taken relative to that clock). From (5) it is easy to see
that a comparison of two different clocks A and B at the same location
would measure variations in their frequency ratio that depend upon
gravitational potential according to
(6)
where a A and a 8 are a-parameters for each clock type, and ( vA /v8 ) 0 is the
constant frequency ratio at some fiducial spacetime location from which U is
measu red.
A null redshift experiment of this type was performed in April 1978 at
Stanford University. The rates of two hydrogen-maser clocks and of an
ensemble of three SCSO clocks were compared over a 10-day period.
During this period, the solar potential U/c 2 changed sinusoidally with a 24-
hou r period by 3 x 10 - 1 3 because of the Earth's rotation, and changed
linearly at 3 x 10 - 1 2 per day because the Earth is 90° from perihelion in
April . However, analysis of the data revealed no variations of either type
within experimental errors, leading to a limit on the LPI violation
94 C. M. Will
parameter l cx H - cx scso l < 2 x 10 - 2 (Turneaure et al. , 1983).

The art of atomic timekeeping has advanced to such a state that it is now
routine to take redshift and time-dilation corrections into account in
making comparisons between timekeeping installations at different altitudes
and latitudes, and in navigation systems using Earth-orbiting atomic clocks.
For other tests of LPI , including the constancy of the fundamental non
gravitational constants, see TEGP, section 2.4(c), and UPDATE, section
2.3.4.
Newton's WEP has stood the test of more than 300 years of experimental
scrutiny, without any confirmed indication of a violation. When augmented
by the well-tested principles of LLI and LPI to form EEP, it becomes the
basis for the modern view of gravity as a consequence of cu rved spacetime.
The overwhelming empirical evidence supporting these fundamental
principles has convinced theorists that only metric theories of gravity have a
hope of being viable. As a consequence, most of the discussion of
experimental gravitation centers on metric theories. However, before we
deal with those tests, we must consider several more fundamental concepts
that are common to Newtonian gravity and general relativity.
5 .3 .2 The inverse square law

Although the inverse square law is one of the centerpieces of Newtonian
gravitation , the Principia does not contain a general statement of the law ,
analogous, say, to the statement of the three laws of motion that appear in
the section headed 'Axioms, or Laws of Motion'. The closest Newton comes
to a 'law' of gravity occurs in Book III , in the General Scholium to
Proposition XLII, where he states that gravity operates 'according to the
quantity of the solid matter which they [the sun and planets] contain , and
propagates its virtue on all sides to immense distances, decreasing always as
the inverse square of the distances' . Apart from the fact that this law
accounted for the motions of the planets, Newton gave a direct
demonstration of its validity by showing, in Book III , Proposition IV, that
the force required to keep the Moon in its orbit is related to the force of
gravity on the surface of the Earth by the inverse square of the Moon's
orbital radius in units of the Earth's radius.
As we shall see in Section 5 4, up to the accuracy at which post-Newtonian
corrections play a role, the inverse square law of Newtonian gravity has
been verified to high accuracy in planetary and lunar motion . Observations
of planetary, solar, and stellar structure support the law as applied to bulk
matter. Further agreement has been found for determinations of the Earth's
gravitational field via comparisons of surface, rocket, and satellite

measurements, and via comparison of laser ranging to the Moon and to the
LAGEOS satellite. These observations confirm the inverse square law for
distances typically from 1 km to several astronomical units.
However, at shorter distances, the inverse square law has recently come
under more intense theoretical and experimental scrutiny. One reason for
this interest is the possibility that examination of the inverse square law
could reveal information about theories of the strong, weak,
electromagnetic, and gravitational interactions. In some models, exchange
of various types of particles of small but non-zero mass (axions,
hyperphotons, etc.) can lead to a Yukawa-type contribution to the
interaction potential between bodies that could be of sufficiently long range
to be observable. In some models, the resulting force can be combined with
the Newtonian gravitational force, and can be written
F = - G(r)m 1 m 2/r 2 . (7)
where m 1 and m 2 are the masses of the bodies, r is their separation, and G(r)
is an 'effective' gravitational constant, given by
G(r) = G 00 [ l + et ( l + r/A) e - rf).J , (8)
where a characterizes the strength of the coupling of the massive particle,
and A = h/mc is its Compton wavelength.
Despite an early report of a deviation from the inverse square law in the
range of 5-30 cm , later experiments ruled this out. The methods in these
laboratory inverse square experiments range from variants of the Cavendish
experiment using various massive sources (rings, cylinders, oil tanks, etc.) to
direct measurements of the divergence of g, which vanishes if the inverse
square law is valid (see Section 5.2), using an array of orthogonal gravity
gradiometers. (For reviews and references see Newman ( 1983), UPDATE,
section 2 . l (iv), Hoskins et al. ( 1985), and Stacey et al. ( 1986).) However, in
the range 1- 1000 m, measurements of the gravity profile in deep mines have
yielded values of the gravitational constant at depth that appear to differ
from the laboratory values obtained from Cavendish-type experiments. In
these experiments, measurements of the local acceleration g and its gradient
at different depths in mines are combined with determinations of the local
matter density from core samples, in order to take into account deviations in
density from that of an idealized Earth model. The resulting values of G are
between 0.5 and 1 per cent higher than the laboratory values. One source of
systematic error in such experiments is the possibility of unsampled density
anomalies. Other measurements of this type involve determining the
96 C. M. Will
gravitational acceleration at various depths in large bodies of water, such as

the oceans or reservoirs (Holding et al. , 1986; Stacey et al. , 1986). Fig. 5 . 1
shows the current limits on the parameters ri and A. i n (8) that ar.e imposed by
the best current experiments.
5 .3 .3 Does action equal reaction?

Perhaps the most original and far reaching of Newton's laws of motion is the
third law : 'to every action there is always opposed an equal reaction : or the
mutual actions of two bodies upon each other are always equal, and directed
to contrary parts'. (The first two laws are the law of inertia and the force
law.) This law allowed Newton to free himself from the constraint of treating
gravitation as strictly a one-body problem, and to deal with two or more
Fig. 5. 1 . Limits on coupling strength a and range A. of a hypothetical short

range component of gravitational force. Hatched regions are disallowed at
lu level. Constraint in the range 10- 3- 1 0 m comes from laboratory
experiments (Hoskins et al., 1985); constraint in the range > 10 2 m comes
from satellite data. Shaded region in a < 0 portion of figure shows values of a
and A. implied by mine measurements (Stacey et al. , 1 986).
-1
es
-2
.2
-3
-4
-2 -1 log >.. (m) 2 3 4
-4
es
,-.. -3
I
'-'
tlO
..9 -2
-1
0
bodies in mutual gravitational attraction moving about their common

center of mass. But this law goes even deeper, for it embodies the concept of
the conservation of momentum for isolated gravitating systems : thus, as
Newton states in Corollary IV to the Laws of· Motion, 'the common centre
of gravity of all bodies acting upon each other (excluding external actions
and impediments) is either at rest , or moves uniformly in a right line.'
Conservation of momentum is of course one of the cornerstones of
modern physical theory ; it is normally built into the structure of a given
theory by basing the theory on an action principle. Then , according to
Noether's theorem, conservation laws arise naturally in such theories when
the action embodies a symmetry, in the case of momentum conservation, a
translational symmetry. Newtonian gravitation theory can be derived from
an action , as can general relativity and many other theories of gravitation ,
so conservation of momentum is a natural feature. It is remarkable that,
without the benefit of modern concepts of action and symmetry, Newton
still grasped the meaning of conservation of momentum (it is implicit in the
first law of motion) and built it into gravitation through the third law of
motion .
Although there is ample experimental evidence for conservation of
momentum in atomic and high-energy physics, there is rather less direct
evidence of it in specifically gravitational situations. An alternative
statement of Newton 's third law for gravitating systems is that the 'active
gravitational mass', that is, the mass that determines the gravitational
potential exhibited by a body, should equal the 'passive gravitational mass',
the mass that determines the force on a body in a gravitational field. Such an
equality guarantees the equality of action and reaction and of conservation
of momentum, at least in the Newtonian limit.
A classic test of Newton's third law for gravitating systems was carried out
by Kreuzer ( 1 968). Kreuzer's experiment used a Cavendish balance to
compare the Newtonian gravitational force generated by a cylinder of
Teflon (76 per cent fluorine by weight) with the force generated by that
amount of a liquid mixture of trichloroethylene and dibromomethane (74
per cent bromine by weight) that had the same passive gravitational mass as
the cylinder (namely, the amount of liquid displaced by the cylinder at
neutral buoyancy). Kreuzer's conclusion was that the forces were the same,
and hence that the ratio of active to passive mass for fluorine and bromine
were the same to 5 parts in 10 5 , that is,
(mA /mp ) Ft - (mA /mp )er
< 5 x 10 - s . (9)
(mA /mp )er
98 C. M. Will
A remarkable planetary test of Newton's third law was recently reported

by Bartlett and van Buren ( 1986). They noted that current understanding of
the structure of the Moon involves an iron-rich, aluminum-poor mantle
whose center of mass is offset about 10 km from the center of mass of an
aluminum-rich, iron-poor crust. The direction of offset is toward the Earth,
about 14° to the east of the Earth-Moon line. Such a model accounts for the
basaltic maria which face the Earth, and the aluminum-rich highlands on
the Moon's far side, and for a 2 km offset between the observed center of
mass and center of figure for the Moon . Because of this asymmetry, a
violation of Newton's third law for aluminum and iron would result in a
momentum non-conserving self-force on the Moon, whose component
along the orbital direction would contribute to the secular acceleration of
the lunar orbit. Improved knowledge of the lunar orbit through lunar laser
ranging, and a better understanding of tidal effects in the Earth-Moon
system (which also contribute to the secular acceleration) through
LAGEOS satellite data, severely limit any anomalous secular acceleration ,
with the resulting limit
(mA /mp ) A1 - (mA /m p ) Fe
< 4 x 10 - 1 2 . ( 10)
(m A /mp ) Fe
According to "the PPN formalism, in a theory of gravity that violates
conservation of momentum, the electrostatic binding energy Ee of an atomic
nucleus could make a contribution to the ratio of active to passive mass of
the form
( 1 1)
where ( 3 is a PPN 'conservation law' parameter, whose value is zero in any
conservative theory. The resulting limits on ( 3 from the Kreuzer and lunar
experiments are IC 3 1 < 6 x 10 - 2 and IC 3 1 < 1 x 10 - 8 , respectively . Other tests
of conservation laws in post-Newtonian gravity are described in TEGP,
section 9 .3.
5 .3 .4 Is Newton's gravitational constant constant?

It is interesting to note that the term 'gravitational constant' never occurs in
the Principia. In fact, it seems that the universal constant of proportionality
that we now call G, that relates the gravitational force between two bodies to
the product of their masses and the inverse square of their separations, does
not make an appearance until well into the eighteenth century, in Laplace's
M ecanique Celeste, for example, and later in Poisson's Traite de M ecanique.
The reason is that, in applying Newtonian theory to planetary motions, the
absolute value of G is irrelevant, since it is only the product of G with the

unknown solar or planetary mass M that is important, thus the entire
discussion of celestial mechanics can be carried out in terms of ratios
referred to the Sun or the Earth, as it is in the Principia. Even when Bouguer
in the 1730s carried out lab oratory-scale gravitational experiments using
pendula and plumb lines to measure the variation of gravity with altitude
and the gravitational effect of mountains, the goal was not to measure G
directly, but to determine the mean density of the Earth. Similarly, the
classic torsion-pendulum experiments performed by Cavendish around
1797 make no mention of G ; they too were designed to measure the Earth's
mean density. Even though G could in principle be calculated from the
density, the local gravitational acceleration g, and the radius of the Earth, it
was 1 890 before experimentalists such as Poynting, Boys, and others began
to quote absolute values for G, and to regard it as a fundamental constant,
for which a standard value could be determined. t
Whether the gravitational constant was discussed explicitly or not, one
theme that ran as a thread through all gravitational studies up to the
twentieth century was the idea that gravity is unchanging with time, in other
words, that G is constant in time. Such a notion meshed with the prevalent
world view that held that the universe outside the solar system was static,
therefore there was no mechanism for G to vary, no natu ral timescale for a
variation.
However, the discovery that the universe is expanding generated
numerous proposals for a gravitational constant that could vary as the
universe evolves, ranging from the 'large numbers hypothesis' of Dirac in the
1930s to the Machian viewpoint of Dicke and the scalar-tensor theory in the
1960s (see Wesson ( 1978) for a review). In most of these models, the value of
G was viewed as being modulated by a coupling between gravity and some
additional field of cosmological origin. If G does change with cosmic
evolution , its rate of variation should be of the order of the expansion rate of
the universe, i.e., G/G = aH 0, where H 0 is the Hubble expansion parameter
whose value we shall adopt to be H 0 � 55 km s - l Mpc - 1 � (2 x 10 1 yr) - 1 ,
0
and a is a dimensionless parameter whose value depends upon the theory of
gravity under study and upon the detailed cosmological model. In
Newtonian gravity and general relativity, of course, G is precisely constant
(a = O ) .
Several observational constraints have been placed on G/G, usmg
t For an informative history of the Bouguer, Cavendish, and other geophysical
measurements up to 1900, see Mackenzie ( 1 900).
1 00 C. M. Will
methods that include studies of the evolution of the Sun, observations of

lunar occultations (including analyses of ancient eclipse data), planetary
radar-ranging measurements, and lunar laser-ranging measurements. The
present status of these experiments in summarized in Table 5 . 5 (for reviews
of some of these methods see Halpern ( 1978), Gillies ( 1 982) and Reasenberg
( 1 983)). Some authors have claimed that the non-zero results for a shown in
Table 5.5 are significant and support the hypothesis of a varying
gravitational constant, while others have argued that unavoidable errors in
the models and in the different data sets used in the numerical estimation of
parameters such as G/G may seriously degrade such estimates. The most
promising limit or estimate for G/G may come from ranging measurements
to Viking. The combination of three factors : {i) extremely accurate range
measurements made possible by anchoring of the landers and orbiters, (ii)
the unexpectedly long lifetime of the spacecraft (Lander 2 finally died in
November 1982) , and (iii) the ability to combine Viking data consistently
with other data sets such as Mercury and Venus passive radar, M ariner 9
radar and lunar laser-ranging data, has made it possible to look for G/G at
levels below 10 - 1 1 yr - 1 . Current results give upper limits ranging from 3 to
0.6 parts in 10 1 1 yr - 1 . The major factors limiting the accuracy of these
estimates (and . responsible in part for the difference between the two
Table 5.5. Tests of the constancy of the gravitational constant0
cr = (G)
Method x (2 x 101 0 yr) Name
Solar evolution l cr l < 2 Chin and Stothers

Lunar occultations and eclipses lcrl < 0.8 Morrison
<J= - (0.6 ±0.3) Van Flandern
<r= - (0.5 ± 0.3) Muller
<J = - (2.5 ± 0.7) Newton
<r= - (0.3 ±0. 1) Van Flandern
<J < 8
lI
Planetary and spacecraft radar Shapiro et al.
(J < 3 Reasenberg and Shapiro
Anderson et al.
Lunar laser ranging lcr l < 0.6 Williams et al.
Viking radar icrl < 0.6 Reasenberg
<r = 0.04 ± 0.08 Hellings et al.
a For references see UPDATE, section 5.3.

Experimental gravitation from Newton to Einstein 10 1
estimates, despite being based upon similar data sets) are the uncertainty in
the masses and distributions of the asteroids, and the level of correlations
among the many parameters to be estimated in the models. Some authors
have suggested that radar observations of a Mercury orbiter over a two-year
mission (30 cm accuracy in range) could yield �(G/G) -- 3 x 10 - 1 3 yr - 1 (see
UPDATE, section 5.3).
5.4 Testing the applications of gravitation theory

5 .4 . I Motion of the planets
In Proposition XIII of Book III of the Principia, Newton states 'the planets
move in ellipses which have their common focus in the centre of the Sun ; and
by radii drawn to that centre, they describe areas proportional to the times
of description' . This statement was simultaneously a summary of two of
Kepler's three empirical laws of planetary motion, and a statement of the
predictions of the law of gravity. Elsewhere in the Principia, Newton
similarly deals with Kepler's third law, that the square of the period of the
orbit of a planet is proportional to the cube of its semimajor axis. M oreover,
the universality of the proposed law of gravitation meant that the same
statements should apply to the Moon, to the satellites of Jupiter and Saturn,
to the comets, in fact ; to any massive bodies. Thus on the broad level,
Newtonian gravity led to a complete understanding of the principles of
planetary motion. One of the first successes of Newton's theory of
gravitation was its use by Halley to determine that the orbits of the comets
that had appeared in 1 53 1 , 1607 and 1682 were actually the same orbit , and
that the comet should reappear approximately every 75 years. The
reappearance on schedule of Halley's comet in 1986 was a fitting celestial
herald of the tercentenary year.
In reality, of course, the detailed motions of the planets were more
complicated than the simple Keplerian picture, but these could be explained
as being due to the perturbations of a given orbit by the gravitational forces
due to other planets. Thus, for example, Proposition XIV , Book III, states
'the aphelions and nodes of the orbits of the planets are fixed'; nevertheless,
Newton recognized that this is an approximation , and that the inner planets
(Mercury , Venus, Earth and Mars) will experience motions of their aphelia
because of the perturbing effect of Jupiter and Saturn.
Newton could only provide estimates of the effects of perturbations.
During the eighteenth century, such figures as D'Alembert, Clairaut, Euler,
Lagrange, and Laplace brought increasing mathematical sophistication to
1 02 C. M. Will
bear on the problem of celestial mechanics, so that, by 1758, Clairaut was

able to make an accurate prediction of the return of Halley's comet
(confirmed to within one month), Euler and Lagrange were able to
demonstrate that the observed changes in the period of Jupiter and Saturn
were not secular, but were periodic over timescales of centuries, and
Laplace's M ecanique Celeste gave a complete mathematical accounting of
planetary motions. t
The success of Newtonian gravity applied to the planets reached its zenith
in the middle of the nineteenth century, when Adams and Le Verrier
independently predicted the approximate location of a hitherto unknown
planet that had to be present in order to explain anomalies in the motion of
Uranus. Astronomical searches of the predicted location in 1 846 revealed
such a planet : Neptune.
But a cloud soon appeared on the Newtonian horizon with the
announcement by Leverrier in 1859 that, after the perturbing effects of the
planets on Mercury's orbit had been accounted for, and after the effect of the
precession of the equinoxes on the astronomical coordinate system had been
subtracted off, there remained in the data an unexplained advance in the
perihelion of Mercury. The modern value for this discrepancy is 43 arc
seconds per century. A number of ad hoc proposals were made in an attempt
to account for this excess, including, among others, the existence of a new
planet Vulcan near the Sun, a ring of planetoids, a solar quadrupole moment
and a deviation from the inverse square law of gravitation (for a review, see
Roseveare, 1982). Although these proposals could account for the perihelion
advance of Mercury, they either involved objects that were detectable by
direct optical observation yet were not observed, or predicted perturbations
on the other planets (for example, regressions of nodes, changes in orbital
inclinations) that were inconsistent with observations. Thus they were
doomed to failure. General relativity accounted for the anomalous shift in a
natural way without disturbing the agreement with other planetary
observations.
However, since 1967, there has been periodic controversy over whether
the perihelion shift is a confirmation or a refutation of general relativity
because of the apparent existence of a solar quadrupole moment that could
contribute a portion of the observed perihelion shift. To date, the question of
the size of the solar quadrupole moment remains an open one.
The predicted advance, 8.w , per orbit, including both the dominant
t For the early history of celestial mechanics, see Moulton ( 19 14).

relativistic PPN contributions and the Newtonian contribution of the

modified gravitational potential due to a possible oblateness, is given by
L\w = (6nGm/pc 2 ) [-�(2 + 2y - /J) + 1 2 (R 2 c 2 /2Gmp)] , ( 12)
where m = m + m 2 is the total mass of the two-body system ; p = a ( l - e 2 ) is
1
the semilatus rectum of the orbit, with a the semimaj or axis and e the
eccentricity; R is the mean radius of the oblate body ; and 1 2 is a
dimensionless measure of its quadrupole moment , given by 1 2 = (C - A)/
m 1 R 2 , where C and A are the moments of inertia about the body's rotation
and equatorial axes, respectively (for details of the derivation see TEGP,
section 7 .3).
The first term in ( 12) is the classical perihelion shift, which depends upon
the PPN parameters y and {3. (For a summary of the PPN formalism, see the
Appendix .) In general relativity, y = /3 = 1 . The second term depends upon
the solar quadrupole moment 1 2 . For a Sun that rotates uniformly with its
observed surface angular velocity , so that the quadrupole moment is
produced by centrifugal flattening, one may estimate 1 2 to be ,..,,, 1 x 10 - 7 •
Normalizing 1 2 by this value and substituting standard orbital elements and
physical constants for Mercury and the Sun we obtain the rate of perihelion
shift w, in seconds of arc per century ,
w = 42':9 s Ap c 1 ,
-
Ap = [-�(2 + 2y - /3) + 3 x 10 - 4 ( 1 2/ 10 - 7 )] . ( 1 3)
The measured perihelion shift is accurately known : after the effects of the
general precession of the equinoxes (5000" c - 1 ) and the perturbing effects of
the other planets (280" c - 1 from Venus, 150" c - 1 from Jupiter, 100" c - 1
from the rest) have been accounted for, the remaining perihelion shift is
known (i) to a precision of about 1 per cent from optical observations of
M ercury during the past three centuries, and (ii) to about 0.5 per cent from
radar observations during 1966-76. Unfortunately, measurements of the
orbit of M ercury alone are incapable at present of separating the effects of
relativistic gravity and of solar quadrupole moment in the determination of
AP . Thus, in two analyses of radar distance measurements to Mercury, 1 2
was assumed to have a value corresponding to uniform rotation (effect on AP
negligible), and the PPN parameter combination was estimated. The results
were
{ 1 .005 + 0.020 (1966-7 1 data)
31 (2 + 2y - /3) - 1 .003 + 0.005 ( 1966-76 data) ' ( 1 4)
(see TEGP, section 7 .3, for references) where the quoted errors are la
104 C. M. Will
estimates of the realistic error (taking into account possible systematic

errors).
The origin of the uncertainty that has clouded the interpretation of
perihelion-shift measurements is a series of experiments performed in 1 966
by Dicke and Goldenberg. Those experiments measured the visual
oblateness or flattening of the Sun's disk and found a difference between the
apparent polar and equatorial angular radii that was interpreted as
corresponding to a value J 2 = (2.47 4- 0.23) x 10 - 5 . A value of J 2 this large
would have contributed about 3" c - 1 to Mercury's perihelion shift, and thus
would have put general relativity in serious disagreement with the
observations, while on the other hand supporting Brans-Dicke theory with
a value w -::::::. 7 , whose post-Newtonian contribution to the perihelion shift
would thus have been 40" c - 1 .
These results generated considerable controversy within the relativity and
solar physics communities, and a mammoth number of papers was
produced, some in support of solar oblateness, some in opposition . The
controversy abated somewhat in 1 973, when Hill and his collaborators
performed a similar visual oblateness measurement that yielded J 2 =
(0. 10 + 0.43) x 10 - 5 , an upper limit five times smaller than Dicke's value
(TEGP, section 7 .3).
In 1982 the controversy was revived by reported estimates for J 2 as large
as Hill et al.'s previous upper limit, obtained from analyses of the global five
minute oscillations of the Sun. The discovery of these oscillations around
1976 provided a possible new probe of the internal structure of the Sun , in
particular of its internal rotation . In the absence of rotation, the 21 + 1
normal modes of non-radial oscillation corresponding to the spherical
harmonic indices (l, m) are degenerate, i .e., they all have the same
eigenfrequency. However, rotation splits the modes by amounts that depend
on the mode number, on (l, m) and on certain weighted averages of the
angular velocity distribution of the Sun over its interior. Analyses of the
observed splittings by Hill et al. ( 1982, 1986), Gough ( 1 982) , Campbell et al.
( 1983), and Duvall et al. ( 1984) produced estimates for the angular velocity
distribution in the interior of the Sun which were then used in solar models
to obtain values for 1 2 • The results ranged between 10 - 7 and 6 x 10 - 6 .
Meanwhile, new visual oblateness measurements by Dicke and colleagues in
1983 gave a reported value of about 8 x 10 - 6 (Dicke et al. , 1985).
The present situation is illustrated in Fig. 5.2 : plotted are measured values
of J 2 as a function of time. Also shown are the range of values of J 2 expected
in a uniformly rotating solar model and the maximum value of J 2 that would
be compatible with general relativity within the la and 2a upper bounds on

\ obtained from radar observations of Mercury.
One of the major difficulties in relating visual solar oblateness results and
solar oscillation data to J 2 is that a considerable amount of complex solar
physics theory must be employed. There is, however, a way of determining
J 2 unambiguously, namely by probing the solar gravity field at different
distances from the Sun , thereby separating the effects of J 2 from those of
relativistic gravitation through their different radial dependences. One
method would compare the perihelion shifts of different planets. Another
method would take advantage of Mercury's orbital eccentricity ( e 0 . 2 ) and -
Fig. 5 .2 . The problem of J 2. Open circles represent visual oblateness
measurements, filled circles represent solar oscillation inferences. 'General
relativity la (2a)' lines represent maximum values of J 2 that would be
compatible with general relativity within la (2a) errors in radar
determinations of Mercury's perihelion shift. Shaded area represents values
of J 2 that would be expected from a conventional, uniformly rotating solar
model.
Year
1 965 1 970 1 975 1 980 1 98 5
7' Dicke and

Y Goldenberg
Camp bell : Dicke
� �� I Hill
__ mu !
1:
. ·
General relativity 2o _
_ _ _
General relativity 1 o
- - --- -
- - - - - - Gough
Hill
Duvall !
1 0-s
1 06 C. M. Will
search for the different periodic orbital perturbations induced by J 2 and by

relativistic gravity. The accuracy required for such measurements would
necessitate tracking of a spacecraft in orbit around Mercury, but some
studies have shown that J 2 could be determined to within a few parts in 10 7 •
A direct measurement of J 2 could be provided by a mission that was at one
time under study by NASA. Known as Starprobe, it is a spacecraft whose
orbit would take it to within four solar radii of the Sun. Feasibility studies
indicated that J 2 could be measured to an accuracy of 10 per cent of its
conventional value of 10- 7 . Unfortunately, this mission is not a part of
NASA's current plans.
From the time that mathematical techniques for celestial mechanics
became sophisticated enough to carry out detailed perturbation
calculations, the determination of ephemerides of the planets has been
carried out using Newtonian gravity. However, in the past 20 years, the
tremendous improvements in the accu racy of planetary and spacecraft
tracking, and in the ability of theorists to model their motions, especially
using computers, has made it possible to turn from Newtonian ephemerides,
to post-Newtonian ephemerides. This is now the standard approach at
places like the Jet Propulsion Laboratory and the Center for Astrophysics,
where much of the current analysis of relativistic solar-system dynamics is
carried out.
These analyses use a multiparameter least-squares fit of tracking data and
optical observations to a model for the trajectories of the planets and of any
spacecraft of interest, and for the curved-spacetime propagation of the radar
signals (see Section 5 .4.3) . The equations of motion used to model the orbits
are those of the PPN formalism, rather than of Newtonian theory (or even of
general relativity). The parameters that are to be estimated include (i) the
initial positions and velocities of the nine planets and the Moon; (ii) the
masses of the planets, and of the three asteroids Ceres, Pallas and Vesta ; (iii)
the mean densities of 200 of the largest asteroids whose radii are known ; (iv)
the Earth-Moon mass ratio ; (v) the length of the astronomical unit; (vi)
PPN parameters, y, {3, cx 1 , . . ; (vii) J 2 of the Sun ; (viii) other parameters
.
relevant to specific data sets , such as station locations, rotation and libration
of bodies, known systematic errors or corrections, etc. The output of the
model is a set of updated or improved values of the parameters. In addition
to many years of optical observations of the planets, the data set includes
radar-ranging measurements to M ercury, Venus, the Mars orbiter M ariner
9, and the Viking Mars landers, and laser-ranging measurements to the
Moon. Further analyses of post-Newtonian ephemerides will improve our
knowledge of PPN parameters, of J 2 , and of the dynamics of the solar

system. (For further details, see Hellings ( 1 984), Reasenberg ( 1983), and
UPDATE, section 4.4.)
5 . 4 .2 Motion of the Moon

The Moon plays a central role in Newtonian gravitation and in the
Principia. It gave Newton the first test of the validity of the inverse square
law, in his demonstration, carried out at least 20 years before the publication
of the Principia, that the force required to keep the M oon in its orbit, and the
force of gravity at the surface of the Earth were related by the ratio of the
inverse square of the respective distances from the center of the Earth.
Newton withheld announcement of this result until around 168 5 , when he
was fully satisfied with his proof that the field outside a spherical body was
the same as that of a mass point (Cajori , 1928).
The motion of the lunar perigee, caused by the Sun's gravitational
perturbation, was also treated by Newton, incorrectly as it turns out.
Although it is believed that Newton at some point carried out a correct
treatment of the problem, he never incorporated it into any editions of the
Principia, instead the derivation he gave in the Principia leads to a value that
is about half the observed value (Book I , Proposition XLV , Corollary I). In
Book III , he alludes to a correct derivation, but without any calculational
details. The correct derivation is contained in a set of Newton's unpublished
papers now called the Portsmouth collection, but astronomers of the day
were unaware of it. This celebrated discrepancy induced the English
government and several scientific societies across Europe to offer prizes for a
correct calculation of the effect. The problem was finally resolved in 1749 by
Clairaut, and the result was another triumph for Newtonian theory.
Another difficulty in the motion of the Moon was the 'secular
acceleration' , an observed systematic increase in the lunar orbital period.
This problem was attacked by the great mathematicians of the nineteenth
century, such as Euler, Lagrange, and Laplace. These efforts were only
partially successful in accounting for observations, not because of a defect of
Newtonian theory, but because the strong solar perturbation of the Earth
M oon system made it difficult to find a reasonably convergent
approximation method, and because of the complicating dissipative effect of
ocean tides on the Earth. It was late in the nineteenth century before an
accurate and reliable model, the Hill-Brown theory, was devised. The
problem of the secular acceleration was further complicated by the fact that
part of the observed effect is due to the slowing down of the rotation of the
108 C. M. Will
Earth. Although Kant was the first to suggest, in 1754, that the Earth must
be slowing down (arguing that since the M oon has spun down until it
presents the same face to the Earth, presumably from tidal effects, the Earth
must suffer the same slowing down), he later retracted the hypothesis, and it
was not until the work of Delauney in 1866 and Spencer-Jones in 1 939 that
the importance of this effect was appreciated (for a brief review, see the
article by Muller in Halpern ( 1978)).
Today, the agreement of the theory of the Moon's motion with
observations carried out using such modern tools as lunar laser ranging
represents another triumph for Newtonian gravity. Although some workers
have incorporated general relativistic effects into the theory, most of them
are too small to be detectable with current technology (TEGP, section 8.3) .
The effect of tidal dissipation in the Earth on the Moon 's secular
acceleration remains an important uncertainty in any attempts to place tight
limits on a cosmological variation in G (see Section 5.3 .4), but these
uncertainties may soon be reduced significantly through laser ranging to the
Earth-orbiting LAGEOS satellite, from which improved determinations of
the effects of ocean tides can be obtained.
However, there is one way in which the M oon has provided an important
test of relativistic gravity, and it hearkens back to Newton ' s W EP .
According t o Newton, the equivalence principle should apply t o all bodies,
laboratory-sized bodies as well as planets. Indeed, since Newtonian gravity
is a linear theory, it is straightforward to show that the acceleration of a
gravitating body such as a planet in an external gravitational field is the
same as that of a body of negligible size (ignoring tidal effects). But in most
relativistic theories of gravity, including general relativity, gravity is non
linear, so that the internal gravitational field of a massive body can interact
with the external gravitational field, in a manner that could differ from the
interaction of the matter comprising the body. The result would be a
violation of WEP for massive, self-gravitating bodies.
In a pioneering calculation using an early form of the PPN formalism,
Nordtvedt ( 1 968) showed that many metric theories predicted just such a
violation, in other words, they predicted that bodies could fall with different
accelerations depending on their gravitational self-energy. For a spherically
symmetric body, the acceleration from rest in an external gravitational
potential U has the form
a = (mp/m1 ) VU,
mp = m1 - ryEg /c 2 ,
1
rJ = 4 /3 - y - 3 - 3° � - <:x 1 + �<:x 2 - �' 1 -j' 2 , ( 1 5)
where y , /J, e , a 1 , a 2 , ( 1 , and ( 2 are PPN parameters (see Appendix), and Eg is

the absolute value of the gravitational self-energy of the body. This violation
of the massive-body equivalence principle is known as the 'Nordtvedt effect'.
The effect is absent in general relativity (rt = 0), but present in Brans-Dicke
theory (rt = 1/(2 + w )) .
The fact that the Nordtvedt effect is absent in general relativity is part of
what has come to be called the 'strong equivalence principle' {TEGP,
section 3.3). It has also been shown to be valid for the acceleration of highly
relativistic bodies in general relativity, such as neutron stars and black holes.
In other words, all bodies in general relativity fall with the same
acceleration .
The existence of the Nordtvedt effect does not violate the results of
laboratory Eotvos experiments, since for laboratory-sized objects, Eg /m1 ,.....,
10 - 27 , far below the sensitivity of current or future experiments. However,
for astronomical bodies, Eg /m1 may be significant (10- 5 for the Sun , 10 - 8 for
Jupiter, 4.6 x 10 - 1 0 for the Earth, 0.2 x 10 - 1 0 for the M oon), and in fact it is
an 'Eotvos experiment' using solar-system bodies that has yielded a
significant test of the Nordtvedt effect. If the Nordtvedt effect is present
(rt ¥- 0) then the Earth should fall toward the Sun with a slightly different
acceleration than the M oon. This perturbation in the Earth-Moon orbit
leads to a polarization of the orbit that is directed toward the Sun as it moves
around the Earth-Moon system, as seen from Earth , resulting in a
perturbation in the Earth-Moon distance of the form br =
9.2 rt cos(w0 - w5 )t m , where w0 and W5 are the angular frequencies of the
orbits of the Moon and Sun around the Earth (see TEGP, section 8 . 1 , for
detailed derivations).
Since August 1969, when the first successful acquisition was made of a
laser signal reflected from the Apollo 1 1 retroreflector on the Moon, the
lunar laser-ranging experiment has made regular measurements of the
round-trip travel times of laser pulses between McDonald Observatory in
Texas and the lunar retroreflectors, with accuracies in range of 1 5-30 cm.
These measurements were fit using the method of least squares to a
theoretical model for the lunar motion that took into account perturbations
due to the other planets, tidal interactions, and post-Newtonian
gravitational effects. The predicted round-trip travel times between
retroreflector and telescope also took into account the librations of the
Moon , the orientation of the Earth, the location of the observatory and
atmospheric effects on the signal propagation. The 'Nordtvedt' parameter rt
along with several other important parameters of the model were then
estimated in the least-squares method.
1 10 C. M. Will
An important issue in this analysis is whether other perturbations of the

Earth-Moon orbit could mask the Nordtvedt effect. M ost perturbations
produce effects in �r, which, when decomposed into sinusoidal components
occur at frequencies different than that of the Nordtvedt term (e.g., at
angular frequencies w 0 , 2(w 0 - w5 )), and thus can be separated cleanly from
it using the multiyear span of data. However, there is one perturbation, due
to the tidal part of the solar potential , that does have a component at the
frequency (w 0 - w5), with an amplitude of 1 10 km.
Although this term is ten thousand times larger than the nominal
amplitude of the Nordtvedt effect, it turns out , fortunately , that the
parameters that determine its numerical value, such as the Earth-Moon and
Earth-Sun mass ratios, the astronomical unit and the mean Earth-Moon
distance, are known with sufficient accuracy , partly from previous
measurements and partly from the laser-ranging data itself, that this term
can be accounted for to a precision of about 2 cm .
Two independent analyses of the data taken between acquisition and 1975
were carried out, both finding no evidence, within experimental uncertainty,
for the Nordtvedt effect. Their results for Y/ were Y/ = 0.00 + 0.03 (Williams
et al., 1 976) and Y/ = 0.001 + 0.01 5 (Shapiro et al., 1 976), where the quoted
errors are lcr, obtained by estimating the sensitivity of Y/ to possible
systematic errors in the data or in the theoretical model. The formal
statistical errors that emerged from the data analysis were typically much
smaller, of order cr(Y/) FoRM A L + 0.004. This represents a limit on a possible
-
violation of WEP for massive bodies of 7 parts in 10 1 2 (compare Table 5 . 1).

Improvements in the measurement accuracy and in the theoretical
analysis of the lunar motion may tighten this limit by an order of magnitude,
while a comparable test of the Nordtvedt effect may be possible using the
Sun-Mars-Jupiter system (see TEGP, section 8 . 1 , and UPDATE, section
5. 1, for details and references) .
5 .4 .3 Motion of light
The behavior oflight in a gravitational field was not a subject that originated
with Einstein . In fact Newton raised the question in Query I of the Opticks :
'Do not bodies act upon light at a distance, and by their action bend its rays;
and is not this action (coeteris paribus) strongest at the least distance?' In the
late eighteenth century, Michell and Laplace independently considered the
nature of a body for which the escape velocity would equal that of light
.
(what we now call a black hole). Around the same time, Cavendish evidently
calculated the deflection of a light ray that grazes a body, using Newtonian
gravity and the corpuscular theory oflight, as we know from an unpublished

scrap of paper found among his manuscripts (Thorpe, 192 1). The first
published calculation of the Newtonian deflection of light was that of
Soldner in 1803 (see Jaki ( 1978) for a translation and commentary on
Soldner's paper) . In 19 1 1 , unaware of Soldner's work, Einstein used the
principle of equivalence to derive the deflection of light, obtaining the same
value.
In 19 1 5 , with the full equations of general relativity in hand, Einstein
obtained double the Newtonian deflection. The reason is as follows : the
classic derivations that use only the principle of equivalence or Newtonian
theory yield the deflection of light relative to local straight lines, as defined,
for example, by rigid rods. However, general relativity predicts that space is
curved around the Sun, so local straight lines are bent relative to asymptotic
straight lines far from the Sun by just enough to give double the deflection.
The first contribution to the deflection is universal : it is the same in any
theory compatible with the equivalence principle ; the second contribution,
due to space curvature, varies from theory to theory. Within the PPN
formalism, this variation is expressed by the PPN parameter y. Specifically,
a light ray which grazes the Sun is deflected by an amount
68 = 1( 1 + y)(4GM 0 /dc 2 )
= t( l + y) l':7.5, ( 16)
where M 0 is the mass of the Sun, and d is the solar radius (see TEGP, section
7 . 1 , for derivation). Thus measurements of the deflection of light can be
viewed as measurements of the coefficient 1(1 + y).
The prediction of the full bending of light by the Sun was one of the great
successes of Einstein's general relativity. Eddington's confirmation of the
bending of optical starlight observed during a solar eclipse in the first days
following World War I helped make Einstein famous. However, the
experiments of Eddington and his coworkers had only 30 per cent accuracy,
and succeeding experiments were not much better : the results were scattered
between one half and twice the Einstein value, and the accuracies were low
(for a review, see Bertotti et al., 1962). The most recent optical measurement,
during the solar eclipse of 30 June 1973, yielded a value t( l + y) = 0.95 + 0. 1 1.
However, the development of long-baseline radio interferometry has
produced greatly improved determinations of the deflection of light. Long
baseline and very-long-baseline (VLBI) interferometric techniques have the
capability in principle of measuring angular separations and changes in
angles as small as 3 x 10- 4 seconds of arc. Coupled with this technological
112 C. M. Will
advance is a series of heavenly coincidences : each year, groups of strong

quasi stellar radio sources pass very close to the Sun (as seen from the
Earth), including the group 3C273, 3C279, and 3C48 , and the group
0 1 1 1 + 02, 0 1 19 + 1 1 and 0 1 16 + 08 . As the Earth moves in orbit, changing
the lines of sight of the quasars relative to the Sun, the angular separation
between pairs of quasars varies, and the variation can be measured
accurately by the radio interferometer. A number of measurements of this
kind over the period 1 969-75 yielded an accurate determination of the
coefficient !( 1 + y), which has the value unity in general relativity. Their
results are shown in Fig. 5.3.
A recent series of transcontinental and intercontinental interferometric
quasar observations made primarily to monitor the Earth's rotation was
Fig. 5.3. Results of radio-wave deflection measurements 1969-84 (for

references, see TEGP). Horizontal scale at the top represents values of the
PPN parameter combination !( 1 + y) , whose value in general relativity is
unity. Bottom scale gives the value of thi� combination for the corresponding
value of the Brans-Dicke coupling constant w. In the limit w- oo ,
Brans-Dicke theory becomes indistinguishable from general relativity.
Value of t( ] + 'Y)
0.88 0.92 0.96 1 .00 1 .04 1 .08
Radio deflection experiments
Muhleman et al. ( 1 970)

1 969
Seielstad et al. ( 1 970)
Hill ( 1 97 1 )
1 970 Shapiro (quoted in Weinberg, 1 972)

....
Sramek ( 1 97 1 )
=
4)
·�
8. 1 97 1 Sramek ( 1 974)
Riley ( 1 973)
><
4)
....
...
0
"'
1 972 Weiler e t al. ( 1 974)
>
4)
Counselman et al. ( 1 974)
1 973 Weiler et al. ( 1 975) .,
1 974 Fomalont and Sramek ( 1 975) �
1 975 Fomalont and Sramek ( 1 976)
1 980-84 Robertson and Carter ( 1 984)
5 1 0 20 40 00
Value of scalar-tensor w
sensitive to the deflection of light over almost the entire celestial sphere (at
90° from the Sun , the deflection is still 4 milliarcseconds). The data yielded a
value !( 1 + y) = 1 .004 + 0.003, where the error is a formal standard error
(Robertson and Carter, 1984).
One of the major sources of error in these experiments is the solar corona,
which bends radio waves much more strongly than it bent the visible light
rays which Eddington observed. Improvements in dual frequency
techniques improved accuracies by allowing the coronal bending, which
depends on the frequency of the wave, to be measured separately from the
gravitational bending, which does not (for a review, see Fomalont and
Sramek , 1 977).
Measurements of optical light deflection using telescopes in space have
recently come under serious study. Two proposals along this line are (i)
Hipparcos, an astrometry satellite planned for launch in the late 1980s by
the European Space Agency, designed to measure positions of 105 stars to 5
milliarcsecond accuracy (anticipated accuracy in y of 10 - 3); and (ii)
POINTS (precision optical interferometry in space), a proposed orbiting
optical interferometer with microarcsecond accuracy, capable in principle of
detecting second-order or post-post-Newtonian effects in light deflection.
For fu rther details concerning the deflection of light, see TEGP, section 7. 1 ,
and UPDATE, sectibn 4. 1 .
The discovery in 1 979 of the 'double' quasar Q0957 + 56 1 and its
subsequent interpretation as a multiple image of a single quasar caused by
the gravitational lensing effect of an intervening galaxy made the deflection
of light a useful tool in astrophysics and cosmology. The number and
characteristics of images in such lensed systems can be used as a probe of the
mass distribution of the lensing galaxy or cluster of galaxies.
Another important effect of gravity on the propagation of light was
considered neither by Newton nor by Einstein. Instead it was discovered by
Shapiro, in 1964. It is now called the Shapiro time delay, a retardation of
light signals that pass near a massive body, such as the Sun (Shapiro, 1964).
For instance, a radar signal sent across the solar system past the Sun to a
planet or satellite and returned to the Earth suffers an additional non
Newtonian delay in its round-trip travel time, given by
bt :::'. !( l + y ) [240 20 ln(d 2/r)] µs,
- ( 17)
where d is the distance of closest approach of the ray in solar radii , and r is
the distance of the planet or satellite from the Sun , in astronomical units (s�e
TEGP, section 7.2, for detailed derivation and references).
1 14 C. M. Will
In the two decades following Shapiro's discovery of this effect, several

high-precision measurements were made using radar ranging to targets
passing through superior conjunction. Since one does not have access to a
'Newtonian' signal against which to compare the round-trip travel time of
the observed signal, it is necessary to do a differential measurement of the
variations in round-trip travel times as the target passes through superior
conjunction, and to look for the logarithmic behavior of ( 17). In order to do
this accurately, however, one must take into account the variations in
round-trip travel time due to the orbital motion of the target relative to the
Earth. This is done by using radar-ranging (and possibly other) data on the
target taken when it is far from superior conjunction (i .e., when the time
delay term is negligible) to determine an accurate ephemeris for the target,
using the ephemeris to predict the PPN coordinate trajectory near superior
conjunction , then combining that trajectory with the trajectory of the Earth
to determine the Newtonian round-trip time and the logarithmic term in
( 17). The resulting predicted round-trip travel times in terms of the
unknown coefficient !( 1 + y ) are then fit to the measured travel times using
the method of least squares, and an estimate obtained for !( 1 + y). The time
delay is built into current implementations of post-Newtonian ephemerides
(Section 5.4. 1), and is solved for automatically.
Three types of target were used. The first type is a planet, such as M ercury
or Venus, used as a passive reflector of the radar signals ('passive radar').
One of the major difficulties with this method is that the largely unknown
planetary topography can introduce errors in round-trip travel times as
much as 5 µs (i .e. , the subradar point could be a mountain-top or a valley)
that thus introduce errors in both the planetary ephemeris and, more
importantly , in the round-trip travel times at superior conjunction. Several
sophisticated attempts were made to overcome this problem.
The second type of target is an artificial satellite, such as the Mariners 6
and 7 , used as active retransmitters of the radar signals ('active radar'). Here
topography is not an issue, and the on-board transponders permit accurate
determination of the true range to the spacecraft. Unfortunately, spacecraft
suffer random perturbing accelerations from a variety of sources, including
random fluctuations in the fo rces caused by solar wind and solar radiation
pressure , and random forces from on-board attitude-control devices. These
random accelerations can cause the trajectory of the spacecraft near
superior conjunction to differ by as much as 50 m or 0. 1 µs from the
predicted trajectory in an essentially unknown way. Special methods of
analyzing the ranging data were devised to alleviate this problem.
The third target is the result of an attempt to combine the transponding

capabilities of spacecraft with the imperturbable motions of planets by
anchoring satellites to planets. Examples are the Mariner 9 Mars orbiter
and the Viking Mars landers and orbiters.
In all of these cases, as in the radio-wave deflection measurements , the
solar corona causes uncertainties because of its slowing down of the radar
signal ; again, dual frequency ranging helps reduce these errors. Still , the
corona problem was a major factor limiting the accuracy of the most recent
Viking time-delay measurements , despite the use of a dual-frequency
downlink from the orbiters.
The results for the coefficient 1-( 1 + y) of all radar time-delay
measurements performed to 1980 are shown in Fig. 5.4. Analyses of Viking_
data resulted in a 0. 1 per cent measurement (Reasenberg cf al. , i979).
5 .4 .4 Motion of stars in binary systems

Newtonian gravitational theory is a standard tool in the analysis and
understanding of binary and multiple star systems, yet there is no mention in
the Principia of such systems. Newton restricted his attention to the solar
Fig. 5.4. Results of radar time-delay measurements 1968-79 (for references

see TEGP).
Value of to + 'Y)
T T T T T T T I I I I
0.88 0.92 0.96 1 .00 1 .04 1 .08
Time-delay measurements
Passive radar to Mercury and Venus
Shapiro ( 1 968) - - - --<1•a.-------+---1 1
Shapiro et al. ( 1 97 1 ) L
Active radar I
Mariner 6 & 7 Anderson et al. ( 1 97 5 )
Anchored spacecraft
Mariner 9 Anderson et al. ( 1 978)
Reasenberg and Shapiro ( 1 977)
Viking Shapiro et al. ( 1 977) •
Cain et al. ( 1 978)
Reasenberg et al. ( 1 979) --i i- (± 0.00 1 )
.1 .1 .1 .l
5 1 0 20 40 00
Value of scalar-tensor w
116 C. M. Will
system. The reason is that the first 'double star' (conventionally referring to
a pair of stars seen to be less than a few arcseconds apart) was discovered
only around 1650 and , until around 1779, such occurrences were viewed
merely as curiosities caused by the chance juxtaposition of stellar images . In
1767, Michell pointed out that the probability of such a random
juxtaposition was sufficiently low for the closest pairs, that they may 'really
consist of stars placed near together, and under the influence of some general
law'. But in 1779, Mayer suggested the serious possibility of small suns
revolving around larger ones, and Herschel began a systematic search for
double stars. By 1803 , Herschel had demonstrated conclusively that the
changes in relative positions of some of the double star systems that he had
followed over a 25 year period could only be accounted for if the two stars
were 'intimately held together by the bonds of mutual attraction '. The
motions were seen to correspond to elliptical orbits projected onto the plane
of the sky, a confirmation of the Newtonian inverse square law. Between
1827 and 1832, the problem of deriving the orbit from observational data
was solved by Savary, Encke, and Herschel . In 1889, the fi rst of a new class
of binary star systems, the spectroscopic binaries, was discovered . These are
binary systems in which the two stars cannot be resolved telescopically,
instead the varying Doppler shifts in one or both of the spectra are
measured.t
Analysis of binary-system orbits provides information about the masses
of the components, which are basic quantities for the theory of stellar
structure and evolution. Observations of non-Keplerian apsidal advances in
close binaries with tidal interactions can be used to infer something about
the interior density distribution of the components. The evolution of close
binaries containing massive highly-evolved stars undergoing mass transfer
is important in understanding binary X-ray sou rces containing neutron
stars and black holes. Newtonian celestial mechanics is the key tool in all
these orbital analyses. Relativistic gravity has typically played a minor role.
However, there is one system in which relativistic celestial mechanics
must be used instead of Newtonian celestial mechanics : the binary pulsar.
Discovered in the summer of 1974 by Hulse and Taylor, it is a pulsar of
nominal period 59 ms in a close binary system with an as yet unseen
companion (Hulse and Taylor, 1975). From detailed analyses of the arrival
times of pulses (which amount to an integrated version of the Doppler-shift
methods used in spectroscopic binary systems) extremely accurate orbital
t For the history of binary stars, see Aitken ( 1963).

and physical parameters for the system have been obtained (see Table 5.6).
Because the orbit is so close ( � 1 R 0 ) and because there is no evide'nce of an
eclipse of the pulsar signal or of mass transfer from the companion, it is
generally believed that the companion is compact : a white dwarf, a black
hole, or (most likely) a second neutron star. Thus the orbital motion is very
clean, free from tidal or other complicating effects. Furthermore, the data
acquisition is 'clean' in the sense that the observers can keep track of the
pulsar phase with an accuracy of 20 µs, despite gaps of up to six months
between observing sessions. The pulsar has shown no evidence of 'glitches'
in its pulse period.
Three factors make this system an arena where relativistic celestial
mechanics must be used : the relatively large size of relativistic effects [vorb it /
c � (GM/c 2 R)11 2 � 10 - 3J ; the short orbital period (8 hours), allowing
secular effects to build up rapidly ; and the cleanliness of the system, allowing
Table 5 . 6 . Measured parameters of the binary pulsar°
Symbol Value
Parameter (units) (data to Feb . 1985)
(i) 'Physical' parameters
Right ascension a 19h 13m 12�468 ± 0�00 1
Declination {J 16°0 1'08': 16 ± 0·:0.2
0.059 029 995 2709 ± 20
(8 .63 ± 0.02) x 10 - l S
Pulsar period P_P (s)
Derivative of period PP (ss - 1 )
(ii) 'Classical' parameters
Projected semimajor axis aP sin i (light-sec) 2.34 1 85 ± 0.000 12
Eccentricity e 0.6 17 127 ± 0.000 003
Orbital period Pb (s) 27 906.98 1 63 ± 0.000 02
Longitude of periastron w0 (deg) 178.8643 ± 0.0009
Julian ephemeris date of
periastron and reference
time for Pb and w 0 . To 2 442 32 1 .433 2084 ± 0.000 00 12
(iii) 'Relativistic' parameters
Mean rate of periastron
advance < w) (deg yr -
1) 4.2263 ± 0.0003
Gravitational redshift and
time dilation y (s) 0.004 38 ± 0.000 12
Orbital period derivative Pb (ss - 1 ) ( - 2.40 ± 0.09) x 10 - 1 2
Orbital inclination sin i 0.72 ± 0.05
0 Data from Weisberg and Taylor ( 1 984) and Haugan (1986a).

118 C. M. Will
accurate determinations of small effects. Just as Newtonian gravity is used as

a tool for measuring astrophysical parameters of ordinary binary systems,
so general relativity is used as a tool for measuring astrophysical parameters
in the binary pulsar.
The observational parameters that are obtained from a least-squares
solution of the arrival time data fall into three groups : (i) non-orbital
parameters, such as the pulsar period and its rate of change, and the position
of the pulsar on the sky; (ii) five 'classical' parameters, most closely related
to those appropriate for standard Newtonian systems, such as the
eccentricity and the orbital period ; and (iii) four 'relativistic' parameters.
The four relativistic parameters are (dJ) , the average rate of periastron
advance; y , the amplitude of delays in arrival of pulses caused by the varying
effects of the gravitational redshift and time dilation as the pulsar moves in
its elliptical orbit at varying distances from the companion an � with varying
speeds ; Pb , the rate of change of orbital period ; and sin i, where i is the angle
of inclination of the orbit relative to the plane of the sky.
In general relativity, these parameters can be related to the masses of the
two bodies and to measured 'classical' parameters by the equations
(dJ) = 3G 2 1 3 c - 2 (Pb/2n) - 51 3 ( 1 - e 2 )(mp + mc ) 21 3 ( 18)
y = G 21 3 c - 2 e(Pb/2n) 1 1 3 mc (mp + 2mc )(mp + mc ) - 41 3 ,
()
( 19)
192nG51 3 �
P - 51 3
2 4 e + 96 e 4)mp mc (mp + mc ) 1 3 ,
( 1 - e 2 ) - 71 2 ( 1 + n
Pb = 2 n -1
5c5 2n
(20)
(2 1)
where mp and me denote the pulsar and companion masses, respectively. The
formula for (dJ) ignores possible non-relativistic contributions to the
periastron shift, such as tidally or rotationally induced effects caused by the
companion (for discussion of these effects, see TEGP, section 12. l (c)). The
formula for Pb represents the effect of energy loss through the emission of
gravitational radiation , and makes use of the 'quadrupole formula' of
general relativity (for a survey of the quadrupole approximation for
gravitational radiation, see Schutz and W ill ( 1 987)); it ignores other sources
of energy loss, such as tidal dissipation (see TEGP, section 12. l (f)). The
parameter sin i represents the effect of the Shapiro time delay of the pulsar
signal as it passes the companion.
The current timing model that contains these parameters was developed
by Haugan from earlier treatments by Blandford and Teukolsky and by
Epstein (see Haugan , 1985, for review and references). An alternative timing
model , using a slightly different set of relativistic parameters, devised by
Damour and Deruelle ( 1986), has also been used in the data analysis, with
equivalent results. The values shown in Table 5.6 are from data taken
through February 1985 (Haugan , 1986a); results through August 1983 were
presented by Weisberg and Taylor ( 1984).
The most convenient way to display these results is to plot the constraints
they imply for the two masses mp and me . These are shown in Fig. 5.5. From
(w) and y we obtain the values mp = ( 1 .42 + 0.03)m 0 and me =
( 1 .40 + 0.03)m 0 . Equations (20) and (2 1) then predict values for the
remaining parameters l\ = - 2.403 + 0.002 x 10- 1 2 and sin i = 0.72 + 0.03 ,
in complete agreement with the measured values in Table 5.6. This
consistency is also displayed in Fig . 5.5, in which the regions allowed by the
four measured constraints have a single common overlap. This consistency
provides a test of the assumption that the two bodies behave as 'point'
masses , without complicated tidal effects.
Fig. 5.5 . Curves showing constraints on the mass of pulsar and its
companion provided by measured values and estimated errors of the
parameters (w) , y, I\ , and sin i. Uncertainty in (w) is less than width of
sloping straight line. All four constraints overlap in the region near 1 .4 solar
masses for each body.
3 .--��.--�.� -
.. ... �...-��..--��..--��.-�-----,
.
.
·. ·
c:
0
-�Q,
e
0
u
.....
0
. . . :: : . . . .
. .... . .. . .
···
··
..
··
· ··
Q L._��L._��L-��-'--��...._�.L-�.-..__..
0 2 3
Mass o f pulsar (solar masses)
1 20 C. M. Will
One of the most important results of the analysis of the binary pulsar is the
confirmation of the existence of gravitational radiation through the loss of
orbital energy. Although it is normally stated that Newtonian gravity does
not admit gravitational radiation because the interaction is instantaneous,
'
this was not necessarily Newt on's intent. In the gravitational calculations in
the Principia the interaction was assumed to be instantaneous as a necessary
approximation , while in other writings Newton made it clear that he had in
mind a propagating action , in which the ether between two moving,
gravitating bodies might 'begin to rarify'.t In an unsuccessful attempt to
explain the secular acceleration of the Moon , Laplace assumed that the
speed of gravitational propagation was seven million times that of light.
Nevertheless, the simplest mathematical formulation of Newtonian gravity
assumes instantaneous gravitational interactions. On the other hand ,
gravitational radiation is a necessary outcome of general relativity, since the
theory is compatible with special relativity, and with a limiting speed for all
interactions.
By the same token , alternative metric theories of gravity , because of their
compatibility with Lorentz invariance at some level , also predict
gravitational radiation. However, most such theories predict, in addition to
a contribution from 'quadrupole' gravitational radiation analogous to that
of general relativity, a contribution of 'dipole' gravitational radiation. In
such cases, the amount of orbital period decay is expected to be in significant
disagreement with the constraints imposed on the masses by the other three
relativistic parameters. This difficulty dealt a mortal blow to the ' Rosen
bimetric theory', for example (see TEGP , section 12.3, for discussion).
The binary pulsar has yielded a remarkably rich test of gravitation theory .
confirming Newtonian and general relativistic celestial mechanics, and the
existence of gravitational radiation. It has also demonstrated a role for
general relativity in the determination of astrophysical parameters, such as
the mass of a neutron star. The first announcement of the observation of the
orbital period decrease came in late 1978 , just at the beginning of the
Einstein centenary year, and toward the end of the 'decades for testing
general relativity' . It is now appropriate to summarize and to attempt to
look to the future of experimental gravitation .
5.5 Beyond the tercentenary

On the occasion of the tercentenary of Newton's Principia, we find Einstein's
t Letter from Newton to Boyle, quoted by Cajori in Appendix to the Principia .

general relativity in a situation not unlike that of Newtonian theory around

the sesquicentenary. In 1837 , Newtonian gravity was one of the most
successful theories of nature ever formulated . It passed every test with flying
colors , explaining the detailed motions of the planets and comets, including
a host of perturbing effects. It accounted for the motion of the Moon in
detail ; any unresolved discrepancies simply awaited the advent of more
sophisticated approximation methods. Its realm of validity was extended far
beyond the solar system, with its successful application to binary star
systems. Within a decade , its crowning achievement , the discovery of
Neptune, would be accomplished .
However, a change was soon in the air. In 1 843 , the quest for higher
observational and calculational accuracy led Le Verrier to discover the
problem of Mercury's perihelion, although he did not announce it publicly
for another 16 years. In addition, a revolution was underway in another area
of physics. The new understanding of electromagnetism brought about by
Faraday and Maxwell between 1830 and 1 870 led to the new world view of
special relativity, and Einstein's quest for the unification of gravity with
special relativity led to general relativity and to the overthrow of Newtonian
gravity.
In 1987, general relativity is also viewed as one of the most successful
theories of all time. It, too , has passed every experimental test. In a similar
quest for higher accuracy, theorists and experimentalists are contemplating
and designing the next generation of experiments to look for possible
deviations from general relativity, although we see no indication of serious
deviations on the horizon. A similar revolution is underway in elementary
particle physics, and the new understanding of gauge theories of the non
gra vitational interactions has spawned the current quest for a unification of
all the interactions.
Will these modern developments lead to an overthrow of general
relativity and to new gravitational experiments? Most observers believe that
the current attempts at unification , if successful , will not lead to observable
effects at the level of macroscopic physics or at the energies accessible for the
foreseeable future. This superunification, unlike that of Einstein, will not
overthrow the standard, macroscopic, or 'low energy' version of general
relativity. Instead, any modifications are expected to occur at the 'Planck
energy' appropriate to the very early universe, or at singularities inside black
holes. Consequently, gravitation experimenters of the coming decades must
content themselves with testing standard general relativity, in new regimes
1 22 C. M. Will
and to higher orders perhaps, but with little reason to anticipate a failure of
Einstein's theory.
On the other hand, although physics is normally concerned with making
predictions about the behavior of systems, physicists are notoriously poor at
predicting the future of their own subject. So even after 300 years of
experimental gravitation, who is to say that there will not be surprises
around the corner?
Acknowledgement
This work was supported in part by the National Science Foundation
[PHY85- 1 3953J .
Appendix : The parametrized post-Newtonian formalism

The comparison of metric theories of gravity with each other and with
experiment becomes particularly simple when one takes the slow-motion ,
weak-field limit. This approximation, known as the post-Newtonian limit, i s
sufficiently accurate t o encompass most solar-system tests that can be
performed in the foreseeable future. It turns out that , in this limit , the
spacetime metric gµv predicted by nearly every metric theory of gravity has
the same structure. It can be written as an expansion about tqe Minkowski
metric [17µv = diag( - 1 , 1 , 1 , 1)] in terms of dimensionless gravitational
potentials of varying degrees of smallness. These potentials are constructed
from the matter variables (Table 5.7) and from the Newtonian gravitational
potential
f
U(x , t) = p(x' , tJ l x - x' l - 1 d 3 x' .
The 'order of smallness' is determined according to the rules U ,..., v 2 ,..., TI ,..., p/
i
p ,..., 0(2), v ,..., ld/dtl/ld/dxl ,..., 0(1), and so on. A consistent post-Newtonian
limit requires determination of g 00 correct through 0(4), g 0 i through 0(3)
and gii through 0(2) (for details see TEGP , section 4. 1). The only way that
one metric theory differs from another is in the numerical values of the
coefficients that appear in front of the metric potentials. The parametrized
post-Newtonian (PPN) formalism inserts parameters in place of these
coefficients, parameters whose values depend on the theory under study. In
the standard version of the PPN formalism, summarized in Table 5.7, ten
parameters are used , chosen in such a manner that they measure or indicate
general properties of metric theories of gravity (Table 5.8). The parameters y
and /3 are the usual Eddington-Robertson-Schiff parameters used to
Table 5. 7. The parametrized post-Newtonian formalism
A . Coordinate system:
coordinates are (t , x1, x 2 , x 3).

The framework u ses a nearly globally Lorentz coordinate system in which the
Three-dimensional , Euclidean vector notation is used
throughout. All coordinate arbitrariness ('gauge freedom') has been removed by
specialization of the coordinates to the standard PPN gauge (TEG P , section 4.2).
Units are chosen so that G = c = 1 .
B . Ma tter variables:
1. p = density of rest mass as measured in a local freely falling frame momentarily
comoving with the gravitating matter.
2. vi = (x )
d i/d t = coordinate velocity of the matter.
3. wi = coordinate velocity of PPN coordinate system relative to the mean rest

frame of the universe.
4. p = pressure as measu red in a local freely falling frame momentarily comoving

with the matter.
5. TI = internal energy per unit rest mass. It includes all forms of non-rest-mass ,
non-gravitational energy - e.g . , energy o f compression and thermal energy.
PPN
C.
y, /3, � .ct ct 2 , ct 3 , ( 2 . ( 3 ,
parameters:
1, (1 , (4.
E.
p' 3x' ' Uij - p'(x -I x'Mx-x' ) i
Metric potentials:
u
- JIx-x 'I d
J x-x'13 d 3x'
� - Jp'lxp"(x-x') . x'-x" x - x" d 3x'd3x"
w- - x'l3 ( lx - x"I Ix' - x"I )
p'[v' · (x -x'3 )] 2 3x,
A=
I p l1x2 -x'l d
p' ' 'I d 3x

-
IIx x' I
I
3 xv
, d �I
2 I x-xU
I = I
�1 =
continued
1 24 C. M. Will
Table 5.7 (cont.)
E. Metric potentials:
: ' TI ' 3
<1>3 = 1 - x ' I d x' '
I ' v' 3 ,
P
JI; = I j x - x' I
i d x'
F. Stress-energy tensor (perfect fluid)

r 00 = p (l + TI + v 2 + 2 U)
yoi = p ( l + TI + v2 + 2 U + p/p)vi
I;i = pviJ( l + TI + v 2 + 2U + p/p) + p1Yi( l - 2y U)
G . Equations of motion
1. Stressed matter, yµ v ;v = Q
2. Test bodies, d2xµ/dA.2 + rµv;.(dx•/dA.)(dx)./dA.) = 0
3. Maxwell's equations, Fµ v ; v = 41tJµ , Fµ v = Av; µ - A µ ; v
Table 5.8. The PPN parameters and their significancea
Value Value
. .
Value in m semi- in fully-
What it measures relative general conservative conservative
Parameter to general relativity relativity theories theories
y How much space curvatu re 1 y y

is produced by unit
rest-mass?
/3 How much 'non-linearity' /3 /3

is there in the superposition
law for gravity?
Are there preferred-l ocation 0 �
effects?
IX 1 0 IX 1 0
IX 2 Are there preferred-frame 0 1X 0
IX 3 effects? 2
0 0 0
(1 0 0 0
(2 Is there violation of conser- 0 0 0
(3 vation of total momentum? 0 0 0
(4 0 0 0
aFor a compendium of PPN parameter values in alternative theories together with

derivations, see TEGP, chapter 5 .
describe the 'classical' tests of general relativity ; � is non-zero in any theory

of gravity that predicts preferred-location effects such as a galaxy-induced
anisotropy in the local gravitational constant GL (also called 'Whitehead'
effects) ; a 1 , a 2 , a 3 measure whether or not the . theory predicts post
Newtonian preferred-frame effects (TEGP , section 4.3) ; a 3 , ( 1 , ( 2 , ( 3 , (4
measure whether or not the theory predicts violations of global conservation
laws for total momentum (TEGP , section 4.4). In Table 5.8, we show the
values these parameters take (i) in general relativity, (ii) in any theory of
gravity that possesses conservation laws for total momentum, called
'semiconservative' (any theory that is based on an invariant action principle
is semiconservative), and (iii) in any theory that in addition possesses six
global conservation laws for angular momentum, called 'fully conservative'
(such theories automatically predict no post-Newtonian preferred-frame
effects). Semiconservative theories have five free PPN parameters (y , {3, � , a 1 ,
a 2 ) while fully conservative theories have three (y, {3, �). In Brans-Dicke
theory, y = ( 1 + w)/(2 + w ), where w is the adjustable 'coupling constant' of

the theory, while the other PPN parameters are the same as in general
relativity.
References
Aitken , R. G . ( 1963). The Binary Stars. Dover : New York .

Bartlett, D . F. and van Buren, D . ( 1 986) . Physical Rev. Lett. , 57, 2 1 .
Bertotti , B . , Brill, D . R . and Krotkov, R . ( 1962). I n Gravitation: A n Introduction to
Current Research, ed . L . Witten , pp. 1-48 . Wiley : New York .
Cajori, F . ( 1928). In Sir Isaac Newton 1 72 7-1 92 7 , pp. 127-88. Waverley Press :
Baltimore.
Cajori, F. ( 1934). Sir Isaac Newton's Mathematical Principles of Natural Philosophy and
his System of the World. University of California : Berkeley.
Campbell , L . , McDow, J . C., Moffat, J. W. and Vincent, D. ( 1983). Nature, 305, 508 .
Damour, T. and Deruelle, N . ( 1986). Annales de l'Institut Henri Poincare, 44, 263 .
De Rujula, A . ( 1986). Nature, 323, 760.
Dicke, R. H . , Kuhn, J. R. and Libbrecht, K. G. ( 198 5). Nature, 316, 687 .
Duvall, T. L . , Jr., Dziembowski, W. A., Goode, P . R . , Gough , D . 0 . , Harvey, J . W . and
Liebacher, J . W. ( 1984). Nature, 310, 22.
Fischbach, E., Sudarsky, D., Szafer, A., Talmadge, C. and Aronson , S. H. ( 1986a). Phys.
Rev. Lett., 56, 3.
Fischbach, E . , Sudarsky, D . , Szafer, A., Talmadge, C. and Aronson, S . H . ( 1986b) . Phys.
Rev. Lett. , 56 , 2424, 2426.
Fomalont, E. B. and Sramek, R. A. ( 1975). Astrophys. J., 199, 749-55.
Fomalont, E. B. and Sramek, R. A. ( 1976). Phys. Rev. Lett., 36, 1475-8 .
Fomalont, E. B. and Sramek, R. A. ( 1977). Comments Astrophys., 7, 19.
1 26 C. M. Will
Futamase, T. and Schutz, B. F. ( 1985). Phys. Rev., D32 , 2557-

Gillies, G. T . ( 1982). The Newtonian Gravitational Constant: An Index of Measurements.
Bureau International des Poids et Mesures : Sevres.
Gough, D. 0. ( 1 982). Nature, 298, 334.
Halpern , L., ed. ( 1978). On the Measurement of Cosmological Variations of the
Gravitational Constant. University Presses of Florida : Gainesville.
Haugan, M. P. ( 1 985). Astrophys. J., 296, 1 .
Haugan, M . P . ( 1 986a). I n Fourth Marcel Grossman Meeting on General Relativity, ed .
R. Ruffini. North Holland : Amsterdam.
Haughan, M . P. ( 1986b), in preparation.
Hellings, R. W. ( 1984). In General Relativity and Gravitation, ed . B. Bertotti , F. de Felice
and A . Pascolini, p. 365. Reidel : Dordrecht.
Hill, H. A., Bos, R. J. and Goode, P. R . ( 1982). Phys. Rev. Lett. , 49, 1794.
Hill, H. A., Rabaey, G. R. and Rosenwald, R. D. ( 1986). In Relativity in Celestial
Mechanics and Astrometry, ed . J. Kovalevsky and V. A. Bromberg, pp. 345--54. Reidel :
Dordrecht.
Holding, S. C., Stacey, F. D. and Tuck , G. J. ( 1986). Phys. Rev., D33, 3487.
Hoskins, J. K . , Newman , R . D., Spero, R. and Schultz, J. ( 1985). Phys. Rev., D32, 3084.
Hulse, R. A. and Taylor, J. H . ( 1975). Astrophys. J. Lett. , 195, L5 1 .
Jaki, S . L . ( 1978). Found. Phys., 8 , 927 .
Keyser, P . T., Niebauer, T. and Faller, J. E. ( 1986) . Phys. Rev. Lett., 56, 2425.
Kreuzer, L . B . ( 1968). Phys. Rev., 169, 1007.
Lamoreaux, S. K . , Jacobs, J. P., Heckel, B. R., Raab , F. J. and Fortson , E. N. ( 1986).
Phys. Rev. Lett., 57, 3 125.
Mackenzie, A . S. ( 1900). The Laws of Gravitation. American B ook Co. : New York .
Moulton, F. R . ( 19 14). Celestial Mechanics, 2nd edn. Macmillan : New York .
Neufield, D . A . ( 1986). Phys. Rev. Lett., 56, 2344.
Newman, R. D. ( 1983). In Proceedings of the Third Marcel Grossman Meeting on
General Relativity, ed . H . Ning. North-Holland : Amsterd am .
Nordtvedt , K . , Jr. ( 1968). Phys. Rev., 169, 10 17.
Nussinov, S. ( 1986). Phys. Rev. Lett., 56, 2350.
Prestage, J. D., Bollinger, J. J., ltano, W. M . and Wineland, D. J . ( 1 985). Phys. Rev.
Lett. , 54, 2387 .
Reasenberg, R . D. ( 1983). Phil. Trans. Roy. Soc. (London), 310, 227.
Reasenberg, R . D . , Shapiro, I . I . , MacNeil, P . E . , Goldstein, R . B., Breidenthal, J. C.,
Brenkle, J . P . , Cain , D . L., Kaufman, T. M., Komarek, T . A . and Zygielbaum, A . I.
( 1979). Astrophys. J. Lett. , 234, L2 19.
Robertson , D . S . and Carter, W. E. ( 1984). Nature, 310, 572.
Roseveare, N. T. ( 1982). Mercury's Perihelion from Le Verrier to Einstein. Oxford
University Press : Oxford .
Schutz, B . F. and Will, C. M . ( 1987). In preparation .
Shapiro , I . I . ( 1964). Phys. Rev. Lett., 13, 789.
Shapiro, I. I., Counselman, C. C. III and King, R . W. ( 1976). Phys. Rev. Lett., 36 , 555.
Stacey, F. D . , Tuck, G . J . , Moore, G . I ., Holding, S .C., Goodwin, B . D . and Zhou, R.
( 1986). Rev. Mod: Phys., in press .
Thieberger, P. ( 1986). Phys. Rev. Lett., 56, 2347 .
Thodberg, H . H . ( 1986). Phys. Rev. Lett. , 56, 2423.
Thorpe, E. , ed . ( 192 1). The Scientific Papers of the Honourable Henry Cavendish , F.R.S.,
Vol . II, p. 437. Cambridge University Press : Cambridge.
Turneaure, J. P . , Will, C . M . , Farrell , B. F., Mattison, E. M. and Vessot, R . F. C.

( 1983). Phys. Rev., 027, 1705.
Vessot, R. F. C . , Levine, M. W . , Mattison, E. M . , Blomberg , E. L . , Hoffman, T. E.,
Nystrom , G. U . , Farrel l , B. F . , Decher, R . , Eby, P. B . , Baugher, C. R., Watts, J. W.,
Teuber, D . L . and Wills, F. 0 . ( 1980). Phys. Rev. Lett. , 45, 208 1 .
Weisberg , J . M . and Taylor, J . H . ( 1984). Phys. Rev. Lett., 52, 1348.
Wesson , P. S. ( 1978). Cosmology and Geophysics. Adam H ilger : Bristol.
Will, C . M . ( 198 1). Theory and Experiment in Gravitational Physics. Cambridge
Will , C. M . ( 1984) . Phys. Rep., 1 13, 345 .
Will , C . M . ( 1986) . Was Einstein Right? Basic Books : New York.
Williams, J . G . , Dicke, R . H . , Bender, P . L . , Alley, C. 0 . , Carter, W. E., Currie, D. G.,
Eckhardt, D . H., Faller, J . E., Kaula, W . M., Mulholland, J . D . , Plotkin, H . H.,
Poultney , S . K., Shelus, P . J., Silverberg, E. C., Sinclair, W. S . , Slade, M . A. and
Wilkinson, D. T. ( 1976). Phys. Rev. Lett. , 36 , 55 1 .
Winicour, J . ( 1983). J . Math. Phys., 25, 1 193.
6
The problem of· motion in Newtonian and
Einsteinian gravity
THIBAULT DAMOUR
6.1 Introduction
The problem of motion, i .e. the problem of describing the dynamics of N
gravitationally interacting extended bodies, is the cardinal problem of any
theory of gravity. From the publication of Newton's Principia to the
beginning of the twentieth century this problem has been thoroughly
investigated within the framework of Newton's dynamics and theory of
gravity. This has led to the formulation of many concepts and theoretical
tools which have been applied to other fields in physics. For instance, the
early development of quantum mechanics has been greatly facilitated by the
existence of several structures of classical mechanics : canonical formalism ,
the Hamilton-Jacobi equation, action-angle variables, the eigenvalue
problem of secular perturbations of the solar system . . . . In fact, quantum
mechanics was successfully modelled on the formal structures of classical
mechanics.
In contrast, Einstein's theory of gravity has developed within a
conceptual and mathematical framework which was completely alien to the
Newtonian framework, as is indicated by the name given to it by its founder :
the general theory of relativity ('Die allgemeine Relativitatstheorie',
Einstein, 19 16a). As a consequence, the relationship between Einstein's and
Newton's theories of gravity has been , and still is, very peculiar. On the one
hand, from the technical point of view, the existence of Newton's theory
facilitated the early development of Einstein's theory by suggesting an
approximation method (called 'post-Newtonian') which allowed the
theorists to draw quickly some observational consequences of general
relativity. Indeed, this post-Newtonian approximation method , developed
by Einstein, Droste and De Sitter within one year of the discovery of general
relativity, has led to the predictions of: the relativistic advance of the
The problem of motion in Newtonian and Einsteinian gravity 129
perihelion of planets, the gravitational redshift, the deflection of light , the

relativistic precession of the Moon's orbit , . . . (see the survey of De Sitter,
19 16). On the other hand , as emphasised recently by Eisenstaedt ( 1986,
section 6), the use of this post-Newtonian approximation method has had,
from a conceptual point of view, the adverse side-effect of introducing
implicitly a 'neo-Newtonian' interpretation of general relativity. Indeed,
technically this approximation method follows the Newtonian way of
tackling gravitational problems as closely as possible . But this technical
reduction of Einstein's theory into the Procrustean bed of Newton's theory
surreptitiously entails a corresponding conceptual reduction : the
Einsteinian problem of motion being , in fact, conceived within the
Newtonian framework of an absolute (coordinate) space and an absolute
(coordinate) time. However, some recent developments oblige us to
reconsider the problem of motion within Einstein's theory.
On the one hand , the discovery of the binary pulsar PSR 19 1 3 + 16 by
Hulse and Taylor in 1974, and its continuous observation by Taylor and
coworkers (see references in Weisberg and Taylor, 1984), has led to an
impressively accurate tracking of the absolute orbital motion of a neutron
star in a binary system . Now, a neutron star is so condensed that it creates a
very strong self-gravitational field, so the study of the motion of such a
strongly self-gravitating body poses a new theoretical problem that cannot
be treated by the usual post-Newtonian methods which assume that the
gravitational field is everywhere weak. This has obliged the theorists to
develop new methods for studying the motion of 'condensed bodies' which
draw directly on the physical and mathematical richness of Einstein's
theory.
On the other hand , the extreme accuracy of modern observations of the
motion of the Earth, the Moon and the planets obliges the theorists to
reconsider the traditional (post-Newtonian) way of tackling the relativistic
problem of motion . Indeed, although in principle the post-Newtonian
methods can be used in the solar system because the gravitational field is
everywhere weak and the velocities always small, the way they have been
developed is somewhat unsatisfactory, or at least incomplete.
All this means that it is worth looking back in detail at the foundations of
the problem of motion . The most natural way to do so is to begin by
discussing the Newtonian way of tackling the problem of motion which will
be the subject of the next six sections (Sections 6 .2-6 .7) . Our principal aim in
doing so will be to clarify the basic ideas which allow one to put the
Newtonian problem of motion into a manageable form. In doing so we shall
1 30 T. Damour
go into the details of the way these ideas are implemented in the Newtonian
framework, some of which will seem fairly trivial to the reader; but we think
it is important to go through them for the following reason : precisely
because these details are technically trivial (or too well known) , one is liable
to miss the ideas which underlie them, so that, when tackling the
corresponding relativistic problem, one has the tendency to transplant,
within Einstein's mathematical framework , exactly the same technical
developments, instead of trying to transmute, within Einstein's conceptual
framework , the ideas underlying them.
In Section 6.2 we pose the Newtonian problem of the motion of N
extended bodies in a general form. In Section 6.3 we discuss the splitting of
this problem into two sub-problems : the external problem (the motion of a
body as a whole), and the internal problem (the intrinsic motion of each
body) . This splitting is useful because the two sub-problems are only weakly
coupled. Very fundamental facts underlie the remarkable weakness of this
coupling, and these are discussed in Sections 6.4 and 6.5 . In Section 6.4 we
study one aspect of this weak coupling, namely the remarkable smallness of
the influence of the internal structure on the 'external' motion . Reviving the
terminology introduced by M. Brillouin ( 1922) and T. Levi-Civita ( 1937a),
we shall speak of an 'effacement' of the internal structure in respect of the
external problem. In Section 6.5 we study the other aspect, namely the
remarkable smallness of the influence of the external motion on the internal
one. Extending the use of Brillouin's and Levi-Civita's terminology, we
shall speak of an 'effacement' of the external structure in the internal
problem. These 'effacement' properties are closely linked with the so-called
'principle of equivalence' (in a sense they can be regarded as precise
formulations of some aspects of the principle of equivalence, the latter often
being expressed either in a vague manner or in a precise but restricted one).
While preparing this article I went back to read those sections of the
Principia in which Newton discusses the equivalence of inertial and
gravitational mass. Newton was fully aware of the remarkable character of
this equivalence, and he gives as many arguments as possible in its favour. In
"
particular I found , much to my surprise, that Newton gives a very
interesting argument, about the relative motion of celestial bodies, which
anticipates ideas recently put forward by Nordtvedt ( 1968) within the
framework of relativistic theories of gravity (see Section 6.6). We conclude
our discussion of the Newtonian problem of motion by outlining the way the
two sub-problems can be effectively solved (Section 6.7).
In the second part of this article (Sections 6.8-6. 1 5) we discuss the
Einsteinian problem of motion. I n Section 6.8 we pose this problem i n a

general form . In Section 6.9 we discuss the dimensionless parameters of the
relativistic problem which can be used to concoct various approximati,:m
methods. In particular, we discuss the post-Newtonian methods (Section
6. 10), the post-Minkowskian ones (Section 6. 1 1 ) and some classes �f
'singular perturbation' methods which have been found useful in general
relativity (Section 6. 12). In Section 6. 1 3 we survey the various approaches
which have been taken to try to 'transplant' or to 'transmute', within
Einstein's theory, the classical splitting into external and internal problems.
In particular, we discuss the problem of the definition of a 'centre of mass'
(Section 6 . 1 3 . 1). We then give references to the explicit results which have
been obtained for the external problem (Section 6. 13 .2), and we conclude by
discussing the relativistic internal problem which is still poorly understood
(Section 6. 1 3 .3). In Section 6. 14 we indicate what is known about the
effacement of internal structure in the relativistic external problem. For
pedagogical reasons we give a detailed treatment of another derivation of
the Newtonian effacement property (already discussed in Section 6.4).
Indeed, this new derivation allows one to grasp the physical and
mathematical mechanisms at work in the Einstein-Infeld-Hoffmann
approach, simplified by the use of complex analytic continuation, which has
been recently used to prove the strongest existing result of 'effacement'
within Einstein's theory. Finally, in Section 6 . 1 5 , we present the recent
progress which has been achieved in the relativistic two-body problem, and
we show how the secular acceleration· effect observed in the binary pulsar
can be considered as a relativistic descendant of an effect conceived by
Laplace within a modification of the Newtonian framework. A brief
conclusion ends this article (Section 6 . 16).
As our presentation of the problem of motion within Einstein's theory will
necessarily be incomplete, and will probably also be biased towards the
author's point of view, the reader is urged to consult other more complete,
or differently biased, reviews : e.g. Ehlers ( 1977, 1980), the proceedings of the
67th Enrico Fermi School (Ehlers, 1 979) (especially the reviews of D'Eath,
Dixon and Havas), Walker and Will ( 1980), Will ( 198 1), Cooperstock and
Hobill ( 1 982) , the proceedings of the 1982 Les Houches-Nato Advanced
Study Institute on Gravitational Radiation (Deruelle and Piran, 1983)
(especially the reviews of Thorne, and Damour, and the round table on the
equations of motion reported by Ashtekar), the proceedings of the 10th
international conference on General Relativity and Gravitation (Bertotti et
al., 1984) (especially the reviews of Damou r, and Walker, and the workshop
1 32 T. Damour
report of Ehlers and Walker), Schutz ( 1 984, 1985), Will ( 1 986), and the
proceedings of the I 14th IAU symposium (Kovalevsky and Brumberg,
1 986) (in particular see B romberg, and Grishchuk and Kopej!cin).
6.2 The N-extended-body problem in Newtonian gravity

The problem we have in mind is to describe, as accurately as possible, the
dynamics of a system of separate extended bodies, each one having finite
dimensions and being endowed with some internal structure governed by
some equations of state. We shall first assume that it is possible to deal with a
finite system, say the solar system, a star cluster, or a galaxy, and to consider
it as isolated from the rest of the universe. This assumption will have to be
justified or qualified later, using, on a higher hierarchical level , the same
tools as the ones we are going to review concerning the simplification of the
description of a given finite system : for instance, we would justify the
possibility of considering the solar system as isolated by arguing that the
stars of our Galaxy will cause only negligible tidal forces in the solar system ,
thereby leaving its internal dynamics unaffected, although it could certainly
affect the overall motion of the solar system. However, such hierarchical
reasoning leaves the issue of the global justification of our assumption
unanswered, but this should not worry us too much because such an
assumption is made in nearly all other scientific investigations, and might
even be indispensable for any 'falsifiability' of predictions (in Popper's
sense).
Having thereby restricted our attention to the dynamics of a finite system
of bodies, let us further assume that our bodies are made of some perfect
fluid with a given (isentropic) equation of state linking the pressure, p, to the
mass density p :
p = p (p ) . (1)
However, this assumption i s made only to simplify our presentation and
should be replaced, in each case of interest, by the actual equations of state
governing the internal structure of the constituent bodies (for instance the
Hooke equations for an elastic solid). Then the equations describing the full
Newtonian dynamics of the system are : the equation of continuity
ap a (pvi)
0, (2)
ai- + ax i
where vi(x, t) is the velocity field (in cartesian coordinates xi ; i,j, . . . = 1 , 2, 3);
the Euler equations for the local fluid motion
(3)
where U(x i , t) is the (positive) gravitational potential ; and the Poisson

equation
!1. U = - 4nG p , (4a)
where G denotes Newton's constant (G = (6.672 + 0.004) x 10- s cm 3 g - 1

s - 2 ) . The former assumption of 'isolation' of the finite system under
consideration can be interpreted as meaning that the gravitational potential
U should fall off outside the system,
t=lxlc--+onst.
oc
x
limit ( U( , t)) = o (4b)
(where lxl is the Euclidean norm of x). Then the only acceptable solution of
G Jp ',
(x t)
Poisson's equation is :
. t) =
U(x, d 3 x, , (5)
x -x
I. .,1
Ix x x',
where - ' I denotes the Euclidean distance between the field point and
the source point and d 3 x' denotes the Euclidean volume element in
x
cartesian coordinates, dxi t dx' 2 dx' 3 .
p(x , v(x,
Now the dynamics of the system are described by the evolution in time of
t) and t), the evolution of the latter variables being governed by eqs.
p, as given by eqs. ( 1) and (5). Moreover, the mass density

constrained to have a compact support consisting of N non-overlapping
p (x ,
(2) and (3) in which p and U must be replaced by their expressions in terms of
t) is further
connected components ('N-extended-body problem'). From the

mathematical point of view the latter constraint greatly complicates the
problem of the evolution of the system (2)-(3) . In fact , as far as I know, there
exist no theorems which assure the existence of solutions to this system.
(Even in the absence of gravitational interaction the fact that takes the
value zero outside the bodies creates serious problems when trying to prove
p
the existence of solutions.) From the physical point of view, what
complicates the problem of controlling the evolution of the system (2)-(3) is
that it is both non-linear and non-local . One therefore needs to resort to
approximation methods.
1 34 T. Damour
6.3 The external and the internal problems of motion (Newtonian

case)
Tisserand ( 1960) in his famous Traite de M ecanique Celeste (vol . I , pp. 5 1-2)
clearly spelled out the traditional way of tackling the N-extended-body
problem in Newtonian gravity. It consists of decomposing the problem into
two sub-problems, namely (using the terminology of Fock ( 1959)) :
(i) the external problem: to determine the motion of the centres of mass of
the N bodies;
(ii) the internal problem: to determine the rriotion of each body around its
centre of mass.
Let us label the bodies by the latin letters, a, b, c = 1 , 2, . . . , N. For the
ath body, which occupies the volume �' we define its total mass
I
ma == p (x, t) d 3 x (6)
J v.
(ma being constant thanks to the equation of continuity (2)), and the position
of its centre of mass
(7)
Then the well-known consequence of the continuity equation,

d I . . I dF(x, t)
F( x, t ) p (x, t) d 3 x = p(x, t) d 3x ,
dt J v. J v.
(8)
dt
where
--- :-
dF(x, t) aF(x, t)
.
. aF(x, t)
+ v' (9)
dt at ax1
denotes the 'convective derivative' of F, implies, when applied twice to the
definition (7) ,
i
d 2 zi (' d v
ma la = p d 3 x .
J v. dt
( 10)
dt
i
By the local equations of motion p dvi/dt = !F , where !Fi denotes the local
force density, so that
.
d 2 z�
ma
2 =
dt iv.
:;; 1. d 3x ' ( 1 1)
('centre-of-mass theorem'). In our perfect fluid model , eq. (3), one has
o;;- i ap
= - -. + p -. .
au
.':#' ( 12)
ax1 ax1
Within each body the force density ( 1 2) can be decomposed in an internal

force , or self-force,
a u c s )a
'¥1. op
xj + p � ·
·- - ( 1 3)
. (s)a · - a
and an external force,

a u ce)a
-p
.
ffl e ·· -
oxi '
)
( 14)
( a
of the gravitational potential is (x E �)

xx - x
where the self-part u<s)a
p ( ' , t) d 3 x ' ·
its external part, u ( e )a

'
u 1s Ja(.X t) = = G
fv. 1 ·
'
•11
= u - u rs)a ' resulting from an integration over the
( 15)
other bodies, is
u<eJa(.x , t) : = L G I p.(x ' ,.�) d 3 x ' .
J v,, l x - x I
( 16)
b ol a
In studying the internal motion of the ath body it i s also traditional

(although often implicit) to introduce an accelerated 'centre-of-mass frame
of reference' with axes parallel to the global cartesian axes and with
origin the centre of mass of the ath body. The position of any point, P, with
respect to this frame, Ya • will be linked to its position in the global cartesian
frame, .X, by
y� = x i - z�(t). ( 17)
Time being absolute in Newtonian mechanics, the relative velocity,
. d y�
WaI .· -
_� , ( 18)
and the relative acceleration of the (material) point P will be, respectively,
w� = vi - dz�/dt, and dw�/dt = dv i/dt - d 2 z�/dt 2 .
Using this notation the basic equation of the external problem (eq. ( 1 1))
reads
d 2 Zai _ c;;; i 3
ma dt 2 -
fv.
.'fl' (e Ja d 3 x + !#' (s)a d Ya ,
fv.
c;;; i
(19)
while the basic equation of the ath internal problem (motion in the centre-of
mass frame of the ath body) is (from eqs. (3) and (17)-( 18))
( "' ai
OW
p at + w
j uW
:') ai
a ayf,
) - !#' (s)a + fl' (e)
-
c;;; i
.
i
a d 2 Zai
-p 2
dt
(20)
136 T. Damour
a w� . a w� )
(-
or, equivalently, by using eq. ( 19) :
P
t
.i
a + wa a ,j -
Ya
cE
(s)a - !!_
:fl' i
ma
l
V.
.:fl' i
cE _
cE(ie)a !!_
(s)a d 3 Ya + :fl' ma
l a;;(ieJa
V.
.T
d 3x. (2 1 )
From these equations it is apparent that the external and internal

problems are, a priori, intricately coupled. On the one hand the external
problem, eq. ( 19), contains, as second term in its right-hand side, a term of
purely internal origin, the total self-force :
F (is)a ·- l a;;(is)a
·-
v.
<7
d 3 Ya ' (22)
and, moreover, the first term in its right-hand side is of mixed external
internal origin (as discussed later). On the other hand, the internal problem,
eq. (2 1 ), contains, as third and fourth terms in its right-hand side, terms of
external origin, the 'tidal' force-density :
(23)
where we have introduced the notation

P a Ua ) : = p (za + Ya ), (24)
which will be (implicitly or explicitly) extended to all internal quantities.
6.4 The effacement of internal structure in the external problem

A priori, one would think that the previous decomposition of the problem of
motion into two sub-problems is of no value because the main equations
governing the sub-problems are coupled via the terms (22) and (23).
However, there are two circumstances which strongly diminish the coupling
between the external and internal problems. The first, if one may speak thus
of a basic physical fact, is that the total self-force (22), must vanish on
account of Newton's third law of motion ('action and reaction principle'). In
our explicit perfect-fluid model where the local self-force density is given by
eq. ( 13) so that
(25)
this is easily checked by using Gauss's theorem and the vanishing of the
pressure outside the body, for the first term on the right-hand side of eq . (25),
and by replacing the value ( 1 5) for the self-gravitational potential in the
l
second term , leading to
i a u(s)a 3
F ( s )a = O + P a i d Ya
v.
aYa
� - G fJ/;�·�i1�·) (y: - y;') d'y. d 3 y; = O (26)
because of the antisymmetry of the integrand in Ya and y� .

The second circumstance which strongly reduces the influence of the
internal structure on the external motion is that the various bodies making
up the system are usually widely separated . It is convenient to measure the
strength (in fact the weakness) of the 'geometric coupling' between the N
extended bodies by means of a dimensionless parameter
(27)
where L is the characteristic linear dimension of the bodies and R the

characteristic separation between them. We are here following Fock ( 1959)
in characterising the Newtonian coupling between the bodies by only one
parameter. There are situations where the consideration of more parameters
(such as the ratio between the masses) is called for. Spyrou ( 1978) has
attempted such an improved description by recording, in order of
magnitude, the influence of each body b # a on the motion of the ath one. We
will not try to do so here because later in the article we will be mainly
interested in the interplay between the Newtonian parameter (i:x) and the
Einsteinian ones (/3e , /3i , Y e , yJ.
Assuming from now on that
i:x � l , (28)
we can follow the practice of fluid mechanicians (see, e.g . Lagerstrom et al. ,
1967) and consider that the actual system we are interested in
(corresponding to a particular value, say a 0 , of a ) is embedded in a
(fictitious) sequence of systems parametrised by i:x such that 0 < i:x ::::;; i:x 0 . One
is then looking for solutions of the external problem in the form of
asymptotic expansions when i:x tends to zero .
As a first step towards finding such solutions one in fact starts by
expanding the basic equation of the external problem itself, i .e. eq. ( 19),
which reads, taking into account the vanishing of the self-force (eq . (26)),
and the definition ( 14) of the external force :
(29)
1 38 T. Damour
where, for notational simplicity, we have denoted partial differentiation

by a comma : U,i == a v;ax i. Now, the Taylor expansion of U�7 la (ia + Ya )
about Ya = O,
..
U(�la (ia + ya ) = V',�1 la ( ia ) + Ya
,1
2 ia yka U(,�1Jl.k( ia ) + O (I Y
,J U ,'�>.IJa ( ia) + 1-y a .. a 1 3 ) • (30)
yields the desired asymptotic expansion about = 0 : a
d 2 zi
ma a = ma V'�>
dt 2
__
,1
a (i ) + 1-Jiak U'��
a 2 , 1Jk( ia ) + 0ext (a 3)' (3 1 )
where we have used the vanishing of the first-order relative mass moment
I�(t) = = f d 3 YaPaUa , t)y� = O (32)

J v.
(as deduced from definitions (7) and ( 1 7)), where
I�(t) = = f d 3 YaPa (Ya• t)y�fa (33)

J v.
oexl(a3)
denotes the 'inertia tensor' (second-order relative mass moment) and where
denotes a remainder term (mainly due to the third-order moment
cou piing) which is of the order of mI.J a4 U ,..,,, m(L/R) 3 a U, i .e. smaller than a3
the first term in the right-hand side of eq. (3 1). In other words, if one uses a
system of units adapted to the external problem, i .e. such that a
characteristic mass, m, a characteristic separation , R, and Newton 's
constant (or a characteristic orbital time), are all equal to one :
m= R = G = 1, (34)
a,
then the first term on the right-hand side of eq. (3 1) is numerically of order
one, the second term is numerically � 2 while the remaining term is �
Note that by its definition ( 16) the Laplacian of U (e)a is zero within the ath
a3 .
body so that we can replace, in eq . (3 1), the inertia tensor, I�, by its
(symmetric and) trace-free part , the 'quadrupole tensor' :
3
Q aii ( t) : = fai
and thus the equation for the external motion reads :
a• _ l.b ijJkk (35)
,1 (ia) + 1-2 Q aik U(,1J�>.ak( ia) + 0ex1 (a 3 )

d 2 zi
ma a = ma U(�>a
__
·
(36)
dt 2
If we introduce an 'ellipticity parameter' , t:,
Q
t: := sup I I
a
( II:fl? ) (37)
(the vertical bars denoting the Euclidean norm of tensors), then 0 � t: � 1 and
the magnitude of the second term on the right-hand side of eq . (36) is in fact
� ea 2 , which can be much smaller than a 2 if, as is frequently the case (for
reasons to be discussed later), e � 1 .
Finally, we mu st replace the external potential u<e)a (eq. ( 16)) in eq. (36), i.e.
u<e)a (x , t) = L G I d 3y b f b !b ' �
( t
(38)
b#a J �
'
l x - zb - Ybl
by its a-expansion. The following Taylor expansion in powers of Yb (with
· = = oA/oxi)
( ) ( )
A ,I '
i
i
1 _
1 1 1 1
1 3 (39)
I x. _ z. b _ Y. b1 - 1 x· - z·b I - 1 x· - z· bI ,i Yb + -2 1 x· - z· b I , Ybyf, + O ( I Y. b ) ,
ii
.
nullity of the Laplacian of I x - zb 1 ,

yields, again using the vanishing of the first-order relative moment and the
- 1
u<e )a (x, t) = L { + � + �2 G Q� ( + ) } + o exl (a 3). (40)

Gm 1
- z Ix z bl ,ij
+
b# a Ix bl -
Hence we get for the a-expanded equation of external motion :
d2
� {o 1
ma d t � = L G mamb :i i 1 · · 1
b# a UZa Za - Zb ( )
+ � G (m Q{ + m,Q� ) az: :� az: Cz. � z,1 )} + 0'" (� 3).
, • ·
(4 1)
Now the system of eqs. (4 1), with a = 1 , . . . , N, does not yield an
autonomous system for the external motion because it contains, along with
the variables i0 ( t) describing the external motion, and the constant
parameters ma, the time-varying quadrupole tensors Q�(t) which depend on
the internal motion of the bodies. However, in eq. (4 1) the contribution of
the internal structure Q ii( t) to the net force is only ,..., Gem 2 I.3/R 4 , which is ea 2
smaller than the Newtonian force between two point-masses - Gm 2/R 2
(
(first term on the right-hand side of eq. 4 1)). On the other hand, an a priori
order of magnitude estimate of the influence of the internal structure on the
first form of the equation of the external problem , eq . ( 19), would have
yielded an internal-structure-dependent contribution Gm 2/L2 , possibly ,...,
'
coming from the self-force. Therefore, in Newtonian theory there is a drastic
reduction of the influence of the internal structu re on the external motion by
a factor ,..., ea 4 from a priori estimates, and by a factor ,..., ea 2 in the final result.
Using a terminology introduced by Brillouin ( 1922) and Levi-Civita
( 1937a), and revived recently in a study of the motion of two compact bodies
(Damour, 1983a), we shall speak of an effacement of the internal structure in
140 T. Damour
the external problem. When one can neglect the relative order of magnitude
eoc 2 + oc 3 the 'effacement' allows one to reduce the external problem to the
problem of the motion of N point-masses fully characterised by their
positions, ia(t) , and their (constant) masses, ma , and subject to the
( )
autonomous differential system :
d1zi a l
ma ia = L Gmamb - a i I. - . I . (42)
dt bola Za Za zb
6.5 The effacement of external structure in the internal problem

The exact equation governing the internal motion is (from eq . (2 1), taking
into account the vanishing of the self-force),
( aw�
Pa a t +
�
)
. aw� - -
- Pa, i + Pa u(s)a
iZi
> i + .? ( t)a ' (43)
aY!i
where the 'tidal' force density, �t)a (eqs. ( 14) and (23)), is
o;;- ( • ) ·-
( )a Ya · - Pa LJ(e)a
.? it
• • !!.3._
, i {Za + Ya ) _
ma
f V.
/ ., ) LJ(e)a
d 3 YaPa (Ya , i {Za
•1 )
• + Ya . (44)
A priori, the relative order of magnitude of the 'external ' force � t)a , as
compared with the gravitational self-force density, Pa U��)a - Pa Gm/L2 , would
be thought to be given by the order of magnitude of the terms explicitly
appearing on the right-hand side of eq. (44), which are "" Pa Gm/R2 , so that, a
priori, the influence of the external motion (R(t)) on the internal evolution
would be of relative order (L/R) 2 = oc 2 • However, a 'peculiar circumstance'
(which was raised to the status of a fundamental physical fact by Einstein)
causes a reduction between the two terms of the right-hand side of (44),
leading to a fu rther 'effacement ' of the influence of the external motion on the
internal one. Indeed, using the same Taylor expansion, eq . (30), as in the
external problem and again taking into account the vanishing of the first
order relative moment , eq. (32), we find :
:Fi(t)a ( Ya
· ) - P LJ(ela
[ j
· ) ,J + J.P u(e)a ( · ) ,J .J< lak
a ,ij Za Ya 2 a , ijk Za YaYa - m
( ]
a
+ oint (ocs )
(45)
(where again J�k can be replaced by Q�k ).
The first term in the right-hand side of eq. (45) is now of order oc 3 (without
factors e) as compared with the gravitational self-force, while the second and
third terms are, respectively, of order oc4 and oc 5 . In other words, if one uses a
The problem of motion in Newtonian and Einsteinian gravity 14 1
system of units adapted to the internal problem, i.e. such that a

characteristic mass, m , a characteristic linear dimension , L, and Newton's
constant (or a characteristic internal dynamical time) are all equal to 1 :
m = L= G = 1, (46)
then the three terms in the right-hand side of eq . (45) become numerically
- ix 3 , ix 4 and ix 5 respectively. When one can neglect the relative order of
magnitude ix 3 one can completely 'efface' the external motion and reduce the
internal problem to solving the following autonomous integro-differential
system :
(47)
at + a-1 (p w i ) 0,
op a
0
a a
_
-
(48)
Ya
with some given equation of state (1) and the condition that P a and Pa vanish
outside some compact domain �·
6.6 Newton and the strong principle of equivalence

The 'peculiar circumstance' which gave rise, as discussed in the preceding
section , to a stronger-than-expected effacement of the influence of the
external structure on the internal motion is that the coefficient preceding
U��)a, which measures the response of the matter to an external gravitational
field, i.e. the 'passive gravitational mass density', happens to be equal to the
coefficient in front of the acceleration in the local equation of motion (see the
original form (20)), i .e. the 'inertial mass density'. This equality
p pass grav = p inert , (49)
together with the equality, implied by the principle of action and reaction ,
between the 'passive gravitational mass density' and the 'active' one (which
creates the gravitational field , right-hand side of Poisson's equation (4a)),
ppass grav = p act grav , ( 50)
was also partly responsible for the effacement of the internal structure in the
external motion with reduction to the simpler problem of the motion of N
points endowed with only one parameter each, their masses :
minert = mpass grav = mact grav ·
a a a
(5 1)
Newton was fully aware of the very remarkable, and unexpected,
character of the equality (49) and he spends several pages of the Principia
(Newton , 1947, pp. 4 1 1- 1 3 , book III , proposition VI, theorem VI)
1 42 T. Damour
discussing the experimental evidence suggesting it (the equality (50) being

then deduced from the action and reaction principle in the proposition VII ,
theorem VII). He gives essentially four different arguments in favour of (49) :
(i) one concerning the behaviour of test bodies around the Earth (free fall
and pendulum experiments) ;
(ii) one comparing the acceleration of free-fall at the surface of the Earth
with the (centrifugal) acceleration of the M oon in its motion around the
Earth;
(iii) one comparing among themselves the accelerations of a set of satellites
orbiting a common massive object (using the system of the Jovian
satellites, and the solar planetary system);
(iv) one concerning the internal dynamics of a system of satellites around a
body which itself orbits around the Sun (using mainly the Jovian system
but alluding also to the Saturnian one and to the Earth-Moon system).
The second and third arguments, based on the comparison of several
independent absolute measurements, are of poor accuracy (even nowadays)
and therefore provide only a weak confirmation of the equality (49). By
contrast the first and the fourth arguments are of differential type and
provide, according to Newton , two independent confirmations of the
equality (49) with a relative precision of one-thousarn;lth, i .e.,
3,
lbabl < 10 - (52)
where, for a pair of bodies :
(
l)a b ·, -
)(
�ass grav )
gass grav
_ m �.nert - m giert .
-....,-.- (53)
m m
All textbooks on gravitational theory quote the pendulum experiments of

Newton and their result (52) but, surprisingly, textbooks and workers in the
field seem to have ignored Newton's fourth argument in spite of the fact that ,
according to Newton , this argument leads to the same (remarkably good)
result (52). In order to show Newton's profound premonitory achievement
let us go a little bit into his reasoning.
The basis of Newton's reasoning is contained in corollary VI of his third
law of motion (Newton , 1947, p. 2 1 , axioms, or laws of motion) :
If bodies, moved in any manner among themselves, are u rged in the
direction of parallel lines by equal accelerative forces, they will all
continue to move among themselves, after the same manner as if they
had not been urged by those forces.
This corollary immediately follows the famou s corollary V (foe. cit. p. 20) :
The motions of bodies included in a given space are the same among
themselves, whether that space is at rest, or moves uniformly forwards

in a right line without any circular motion ,
which is Newton's form of the principle of (special) relativity. It seems clear
to me, from the proximity of these two statements, and from the similarity of
their formulations, that Newton guessed corollary VI was a kind of
generalisation of corollary V, i.e. a kind of principle of 'general' relativity (in
Einstein's sense). And , indeed , in all the uses of corollary VI in book III he
applies it to the common accelerations impressed on a system of bodies by a
very distant external body when the equality (49) is assumed to hold. In
other words, if we allow ourselves (although this will certainly displease the
science historians) to reinterpret Newton's thought using modern
terminology, we can say that corollary VI predates the Einsteinian principle
of equivalence which tries to raise the equality (4 9) to the status of a
generalisation of the principle of (special) relativity. Now, the recent
literature on theories of gravity (see especially Will, 1979, 198 1) has
distinguished several formulations of the principle of equivalence. The
strongest formulation ('Strong Equivalence Principle') states essentially :
(i) that the equality (49) between passive-gravitational and inertial masses
holds also for self-gravitating bodies, and
(ii) the outcome of local experiments, including gravitational ones,
involving interactions internal to a system embedded in a homogeneous
external gravitational field are the same as in the absence of external
gravitational fields.
Now, the reasoning used by Newton in his fourth argument can be said to
predate the 'Strong Equivalence Principle'. Indeed, this reasoning runs as
follows (a representing for instance Jupiter, and b one of its satellites) : if c50 b ,
defined by eq. (53), is zero then by corollary VI the (gravitational) internal
dynamics of the system {a, b } would be the same as in the absence of the Sun ;
but if c50 b is other than zero then this will perturb the internal dynamics of the
system {a, b } . Newton estimates that the main consequence ofa non-zero c5a b
will be to make the orbit of b around a eccentric, its centre c being displaced
(towards the Sun if c50 b > 0) by the amount :
c5 lic - io l = -!c511 b · lie - io l . (54)
In modern phraseology one can say that Newton predicts a 'polarisation'
of the orbit of the satellite in the direction of the Sun. Nordtvedt ( 1968) (see
also Will, 198 1 , p. 186) has recently 'rediscovered' such an effect and
proposed to test its existence in the Earth-Moon system using the Lunar
Laser Ranging data. Nordtvedt's intention was, similarly to Newton, to test
144 T. Damour
the 'equivalence princ1pie tor massive bodies'. As we shall discuss later, the
strong form of the effacement of internal structure which holds within
general relativity guarantees that in particular Oab = O (even for strongly self
gravitating bodies), but Nordtvedt pointed out that in other relativistic
theories of gravity oab might become non-zero for self-gravitating bodies.
To be fair it should be added that Ne�ton's estimate (54) (given without
intermediate steps) is incorrect both in magnitude and in sign and that the
first correct calculation of the 'polarisation' of the orbit of a satellite by a
relative acceleration field towards the Sun ('gravitational Stark effect') was
achieved by Nordtvedt ( 1968). However, even if Newton's conclusion (based
on the overestimate (54)) that the observations of the Jovian system (in his
3
time) allow one to conclude that loab l < 10 - (a value consistent with
Newton's best pendulum experiments) is too optimistic, we can still admire
Newton's remarkable insight which led him not only to ask profound
questions about the strong equivalence principle but also to suggest where
to look for a possible astronomical violation of the equality (49).
6.7 Solving the Newtonian problems of motion

It is outside the scope of this article to review, even sketchily , the work which
has been done,_ and which still continues , towards solving the equations of
the external and/or internal problems. The reader is referred to the classic
treatises of Tisserand ( 1960) and Poincare ( 1957), and for more recent
developments to Brouwer and Clemence ( 196 1) and Hagihara ( 1 972) which
contain many references. Jardetzky ( 1 958) is fully devoted to the internal
problem. For general results on dynamical systems see e.g. Arnold ( 1978),
Gallavotti ( 1983) and references therein.
Traditionally, it was thought sufficient to deal both with the reduced
external problem (42) (with complete effacement of the internal structures)
and with a simple model for the internal problem (of the type (47)-(48))
including some account of the coupling with the external motion
('precession of equinoxes', . . . ). However, the tremendous increase in
precision brought about by recent techniques (radar ranging, laser ranging,
VLBI , . . . ) calls for extensive refinement of our solutions of the Newtonian
problems of motion. But basically one can use the traditional splitting into
external and internal problems as a very good zeroth approximation.
Indeed, using the 'external units' (34) for the external problem, and the
'internal units' (46) for the internal one, and denoting symbolically by Xe
(resp . XJ a matrix of variables fully describing the external (resp. internal)
evolution, the structure of the exact problem of motion is :
d Xe
0 = - + Eo (Xe) + ex 2 E ( Xe , X J + ex 3 E 3 (Xe , XJ + · · · , (55a)
d te 2
d�
0 = - + lo ( XJ + ex 3 l 3 (X i , Xe ) + ex4J (X i , X e ) + . .
·, (55b)
d ti 4
where te (resp . tJ is the dimensionless external (resp . internal) time (in eq .
(55b) the matrix Xi describing, a priori, a set of fields, can be infinite) .
Therefore, in principle, one can solve the coupled system (55) by an iteration
process which starts with knowledge of the solution of the uncoupled system
obtained as the ex = 0 limit of ( 55). Note also that under normal conditions
the internal motion will be nearly stationary and , in fact , slowly 'driven' by
external influences, so that one expects Xi to vary with the external time
te = ex 3 f 2 ti . (56)
Using te instead of ti in eq . ( 55b) leads to
0-_ 10(XJ + ex 3 / 2 -
dX i
+ ex 3 J 3 (Xi , Xe ) + . . . , (57)
d te
which , together with (55a), can be iteratively solved starting with a
stationary internal trial solution.
Another way of building a zeroth approximation to the system (55) is to
start with the assumption that the N bodies undergo rigid motions. This
allows one to reduce the system (55) (either taken exactly or truncated at
some finite approximation in ex ) to an autonomous system of 6N differential
equations for the 6N parameters describing the translational and the
rotational degrees of freedom of the N bodies (see e.g. Fock, 1959, and
Dixon , 1 979, for detailed accounts of this approach).
However, the extreme accuracy (sometimes even at the centimetre level)
now needed in both the translational motion of the planets (especially the
Moon and the Earth) and the internal motion of the Earth makes it
necessary to go beyond the approximation of rigid bodies, while still taking
into account the smaller ex-dependent couplings in the system (55). For
recent reviews of the state of the art on the problem of Newtonian motion
the reader is referred to Kovalevsky and Bromberg ( 1986) and references
therein .
6.8 The N-extended-body problem in Einsteinian gravity

By analogy with the Newtonian case one usually restricts attention to the
problem of the gravitational dynamics of a finite system of extended bodies.
As recalled in Section 6 .2, the assumption can, more or less, be checked a
1 46 T. Damour
posteriori in Newtonian mechanics by computing the tidal forces due to the

higher hierarchical level of construction of the universe. In J;:instein's theory
the problem is more difficult for two reasons : ( 1 ) as discussed later, one does
not yet have an explicit way of separating the gravitational field, within a
given system, into an internally generated part and an externally generated
one, so that one often relies on Newtonian estimates to justify neglect of the
influence of external matter, and (2) one would like to demand more from
Einstein's theory than from Newton's, namely not only a consistent
hierarchical modelisation of the universe but even a global description of the
universe, because one feels that Einstein's theory has the capacity to yield
one. However, for the time being, a general relativistic global description of
our vast clumpy universe poses a formidable problem (see e.g. Ellis's
discussion, 1984) so that we shall follow the usual approach and limit
ourselves, from the start, to considering an isolated system made of a finite
number of (compactly supported) extended bodies.
As in the Newtonian case (Section 6.2), let us further assume, for the sake
of definiteness, that our bodies are made of some perfect fluid with a given
(isentropic) equation of state linking the proper pressure (measured in a
local rest frame) p 0 , to the proper energy density e0 :
Po = Po (e o ) . (58)
Then the stress-energy-momentum tensor of the fluid reads (in
contra variant form) :
Tµ " = (e0 + p 0 ) u µu " + p0gµ ", (59)
where Greek indices take the values 0, 1 , 2, 3, u" denotes the 4-velocity of the
fluid and gµ)x ;.. ) (inverse of g µ") are the components of the metric tensor in
some coordinate system x µ . Our metric conventions are those of Misner,
Thome and Wheeler ( 1973) (in particular the signature is + + + ).-
Then the equations describing both the dynamics of the matter and the
generation of the 'relativistic gravitational potentials', g µv ' by the matter are
the Einstein equations (c = velocity of light)
8nG
Eµv = -4- I;i v ' (60)
c
where Eµv denotes the Einstein tensor
(6 1)
The Einstein equations (60) replace eqs. (2) (continuity), (3) (Euler) , and
(4a) (Poisson) of the corresponding Newtonian problem. As they are purely
local they must be supplemented by boundary conditions expressing the
assumption that the system is 'isolated'. In contradistinction to the Newton

case where the simple and unique fall-off condition (4b) was sufficient to
determine a physically reasonable solution of the Poisson equation (4a)
uniquely, it is still not known for sure how to formulate supplementary
boundary conditions which ensure the existence and uniqueness of a
physically reasonable solution of the Einstein equations (60). What is
known , however, at least if one stays within the 'isolated system' approach
(see, however, Schutz ( 1 980, 1985) and Futamase ( 1 983) for a different
approach), is that one needs something like a dual-purpose boundary
condition able to express : ( 1 ) that the gravitational field falls off far from the
system, and (2) that there were no 'incoming gravitational waves'
converging towards the system in the remote past. But there is no general
agreement as to how to formulate these conditions in a technically precise
way (see , however, Ehlers ( 1 977, 1980) for the presentation of a tentative but
elegant formulation).
Postponing the enforcement of 'eorrecf'" boundary conditions, we are
faced with the problem ot somehow solving the Einstein equatiofls (60) with
a right-hand side having the structure (59) with the constraint (58) and the
further constraint that c 0 must have as support (in spacetime) N non
intersecting world tubes. This problem is of a higher order of difficulty than
the corresponding Newtonian problem because : ( 1) eqs. (60) are highly m:m
linear in gµ v , and (2) the phenomena of the generation of the gravitational
field by the matter and of the action of the gravitational field on the matter
are intimately coupled in eqs. (60). Indeed , it is well known that the Bianchi
identities imply that a necessary consequence of (60) is
Tµ v ; = 0 ' (62)
v
which, in the case of a perfect fluid (59), yields the relativistic Euler
equations,
(co + Po )UvUµ ; v = - (g µ v + uµ uv )Po ; v , (0-3)
and the relativistic equation of continuity
(r0uµ); µ = 0. (64)
In eq. (64) we have introduced a proper rest-mass density, r0 , defined
(modulo a constant factor) as
r0 = = expf d co
co + Po (co )
. (65)
One also defines a proper specific internal energy density Il0 by

(66)
1 48 T. Damour
At this point one possible way to investigate the relativistic problem of

motion is to postpone the problem of solving the Einstein field equations
(60) (thought to correspond to Poisson's equation (4a )) and to fully exploit
the 'dynamical equations' (62) (which for a perfect fluid yield (63)-(64))
assuming the metric as given. This approach has been developed mainly by
W. G. Dixon , and has led to several remarkably elegant results (see Dixon ,
1979 ; Ehlers and Rudolph, 1977). Recently Bailey and Israel ( 1980) have
succeeded in giving this approach a high degree of formal perfection by
means of a Lagrangian formulation. Moreover, their work leads to the
completion of Dixon 's approach (based on the sole consideration of (62))
with a correspondingly elegant covariant formulation of the field equations
(60). Unfortunately , it seems to me that this formulation (which makes use of
multipolar 'point sources') has only a formal meaning , and that further hard
work is needed to give it a sound mathematical meaning. Moreover, it seems
to me that this whole approach to the dynamics of extended bodies is not
suited to dealing with the motion of strongly self-gravitating bodies which,
however, has become a problem of considerable astrophysical interest.
As we are forced to tackle the Einstein field equations we now ask the
advice of the mathematicians to see how, at least in principle, one can
construct solutions of the partial differential system (60). The most common
approach, due to A. Lichnerowicz and Y. Choquet-Bruhat (see e.g.
Lichnerowicz, 1955; B ruhat, 1 962 ; Choquet-Bruhat and York, 1980 ; York,
1983), starts by considering the system of the field equations (60) together
with the dynamical equations (62). This system is in involution , in the sense
of E. Cartan , and its integration consists of two separate problems : that of
initial conditions and that of time development (or evolution). We shall n ot
enter here into the problem of initial conditions because it is, a priori, not
well suited to consideration of the supplementary boundary conditions
alluded to above (see, however, Schutz ( 1980, 1 985), Futamase ( 1983),
Friedrich ( 198 1 , 1985), Choquet-Bruhat and Christodoulou ( 198 3)). On the
other hand, the constructive method often used by mathematicians to prove
the existence and uniqueness of solutions of the evolution problem can be
used as a guide to tackling the problem of motion. This method consists
essentially in 'relaxing' the field equations (60) (which , a priori, do not form a
partial differential system of well-defined type) by using special coordinate
conditions. The most common conditions (though not the only ones) are the
harmonic coordinate conditions:
(67)
where we have introduced the Gothic contravariant metric variables

gµ v : = Jg g µv , (68a)
g := - d et(gµ v). (68b)
Then the relaxed system governing the evolution of the coupled matter
gravitational field becomes the following partial differential system :
16nG
gaP gµv ,ap + Q µv ( g, o g) = c 4 g { (t: o + Po )uµuv + Po 9µ v } , (69a)
(e o + Po ) U vUµ
·v
'
-
_
( gµ v + Uµ U v)Po · v ,
-
'
(69b)
(r0uµ); µ = 0, (69c)
where Q µ v is a complicated quadratic form in the first derivatives of g .
The problem of the motipn of N extended bodies in general relativity is
then reduced to finding the solutions of the system (69) (with (58) and (65))
which satisfy the 'harmonicity condition' (67), the constraint uµuµ = - 1 ,
some reasonable boundary conditions for 9µv ' and such that the support of
t:0 (and p 0 ) is made of N spatially compact world tubes. This problem is very
difficult, on the one hand because of its strong non-linearity (both in 9µ v and
in uµ) and, on the other hand , because the condition on the support of t:0
makes eqs. (69b, c) non-hyperbolic, so that it is not known , mathematically,
if they admit any solutions at all . One therefore needs, for even stronger
reasons than in the Newtonian case, to resort to approximation methods.
6.9 Approximation methods

The first step towards setting up approximation methods consists of
identifying some small (dimensionless) parameters . We shall assume, as in
the Newtonian case (Section 6.3), that the N bodies making up the system
are widely separated so that the 'geometric coupling' parameter,
L
r:x. := (70)
R'
measuring the ratio between a characteristic linear dimension, L, of the
bodies and a characteristic separation, R, between them is much smaller
than one.
In Einsteinian gravity there exist other parameters which may be small.
Let us introduce f3e : the ratio between a characteristic 'external' velocity
(orbital velocity) and the velocity of light.
vext
/3e : = - . (7 1)
c
150 T. Damour
Similarly, let {Ji denote the ratio between a characteristic 'internal' velocity
(velocity of spinning motion) and the velocity of light :
(72)
If m denotes a characteristic mass let us also introduce two other relativistic
dimensionless parameters, an 'external ' one,
(73)
and an 'internal' one

(74)
Together with the previously introduced Newtonian 'ellipticity' parameter
(eq . (37)),
.
· · ) , (75)
(quadrupole moment) .
t: - ,... .., from sp h enc1ty
·
2 (relative d ev1ation
mL ·
this makes a total of six parameters (rt, f3e , {Ji , Y e , Y i , t:) that can be used to
concoct some approximation methods.
First let us note that' some of these parameters are linked by equalities or
inequalities. Indeed , first ,
(76)
and as it is generally expected that L ;;;;: Gm/c 2 (the equality, in order of
magnitude, being reached only for 'condensed bodies', i .e. neutron stars or
black holes) we shall always have at least
Yi ;$ 1 . (77)
Hence, from our previous minimal assumption about the relative separation
of the bodies we shall have, in general :
Y e :$ rt � 1 (minimal assumption). (78)
On the other hand , we know, a priori, only that f3e :$ 1 , {Ji ;$ 1 , 'Yi ;$ 1 and t: ;$ 1 .
However, if we further assume that the system is gravitationally bound, so
that by the virial theorem (vext ) 2 - Gm/R (or nearly gravitationally bound as
in wide-angle scattering) then we shall have,
p; - Y e � l
(gravitationally bound or wide-angle scattering) , ( 7 9 a )
but in the case of small-angle gravitational scattering we would have
Y e � p; :$ 1 (small-angle scattering). (79b)
In some special situations one can encounter other small parameters, such
as the ratio between some masses (which plays an important role in the
mechanics of the solar system), but we will discuss only the more general
cases (see e.g. Brumberg, 1986, and references therein for the proposal of a
relativistic approximation method based on such a small parameter m/M).
Before embarking on a quick survey of the main approximation methods
which have been used in the problem of motion, let the reader be warned
that there are many different points of view concerning these methods and
that my viewpoint is probably biased . He is therefore invited to consult
other reviews, e.g. Ehlers et al. ( 1976), Ehlers ( 1977, 1980), Walker ( 1984),
Ehlers and Walker ( 1 984), Schutz ( 1984, 1985), Will ( 1986).
6. 10 The post-Newtonian approximation methods

The best-known and most investigated methods of approximation m
general relativity are based on the assumptions that
sup(/3; , f3f ) ;$ sup(ye , yJ � 1 , (80)
i.e. of an everywhere-weak gravitational field, and of slow motions. From
eq . (76) we see that Y e < Y i so that Y i can be taken as the main relativistic small
parameter of these methods, the (independent) parameter rt being
introduced only at a later stage .
The philosophy of these methods is, so to say, to follow in Newton's
footsteps, i .e. to try to set up an approximation method for Einstein's theory
which follows as closely as possible, both technically and conceptually, the
Newtonian way of tackling gravitational problems. In these methods the
role of Einstein's theory is reduced to that of providing small numerical
corrections to the predictions of Newton's theory. As we shall see later, the
danger of these approaches, rightly called 'Post-Newtonian Approximation'
(PNA) methods, lies not so much in technical errors or ambiguities (which
arise only at a high approximation level , see the discussion of radiation
damping below) , but in their overall conceptual framework which misses the
richness and suppleness of the full Einsteinian theory and which amounts
nearly, as was pointed out recently in a somewhat different context
(Eisenstaedt, 1986, section 6), to a 'neo-Newtonian interpretation' of
Einstein 's theory.
Technically speaking, the post-Newtonian approaches seek to model the
solution of the relativistic system (69) on the solution of the Newtonian
system (2), (3), (4a). For instance, eq . (69c) is rewritten as a Newtonian-type
1 52 T. Damour
equation of continuity :
(8 1)
0 0
where the 'time' t = = x /c , the 'velocity' v i = dxi/d t = cu i/u and ,
0
r* := r0 Jg u , (82)
is a 'coordinate rest-mass density'.
Note that I have intentionally refrained from using the letter p to denote
any of the relativistic densities (of energy or rest-mass, proper or
'coordinate'). Indeed, the fact that many authors use the same letter, p, to
denote different things is a source of confusion (and error) in the literature
on equations of motion and I did not wish to compound the confusion.
Moreover, in keeping with what I said in the preceding paragraph, I think
that this confusion of notions pertaining to different theories is not innocent
so that the use of different letters helps to contrast Einstein's theory with
Newton's.
Now the modelling of the reduced Einstein equations (69a) on the
Poisson equation (4a) is achieved at a price of several Ansatze. First one
writes
(83)
where fµv =fµ = diag( - 1 , + 1 , + 1 , + 1) are the usual components of a 'flat'
v
metric. Then hµv is assumed to be everywhere small ('weak gravitational
field') and admits an asymptotic expansion when Yi � 0 of the type :
hµv (x, t) = yi hf2J(x, t) + y? 1 2 hf;i(.x, t) + yt h f�J(x, t) + · · · + yj 1 2h (n')(x , t) + · · ·
(84)
The necessity to introduce half-integer powers of Y i comes from the basic
assumptions (80) which imply that the spatial part of the 4-velocity uk "' f3e �
yl 12 (they also imply that Il0/c2 ,..., p0/r0c2 "' /3i2 � Yi ). Another basic ingredient
of all PNA methods is to require that
1 8hµv 2
8hµv
/
c& "' 'Yil axk
. (8 5)
It is assumed that the expansion coefficients h(n) are always and everywhere
of order unity (or less), and that their spatial derivatives are everywhere of
order 1/L (or less). Note also that, in general , the asymptotic expansion (84)
is of the generalised type where some dependence on 'Y i is implicitly
contained in each coefficient h fn) · In order to give to (84) the rigorous
meaning of a Poincare-type asymptotic expansion one should first specify
which of the parameters determining the gravitational field are fixed in the
limit Yi � 0. For one possible approach along these lines see Futamase and
Schutz ( 1983).
Then one easily sees that if one uses a system of units adapted to the study
of the considered gravitationally interacting system, i .e. such that
G = m = L = 1, (86)
then the numerical value of the velocity of light in these units is such that
(87)
and all the h<11> s , their time or space derivatives, and all the internal or
external velocities become of numerical order unity (or less) . With this in
mind , one can summarise the post-Newtonian assumptions by saying that
one looks for solutions of the system (69) in the form of formal expansions in
powers of 1/c (and it is convenient to use this language even if one does not
explicitly use the system of units (86)).
Among the field equations (69a) the dominant one is the zero-zero
component which reads, at lowest order:
i i
16nG
� (y i h < 2> ) = c 2 r0 + O(y i /L ) .
oo
(8 8 )
Eq. (88) is very similar to the Poisson equation (4a), but like the latter it
does not determine h?2� unless we supplement it by some boundary
conditions. Many authors working with PNAs have implicitly assumed that
it was sufficient to use the same boundary condition as in the Newtonian
case (eq. (4b)) namely :
limit (hf,�(x, t)) = O. (89)
lxl --+ oc
t = co st
n
At the lowest level of approximation, (88), this uniquely determines (if we

decide to use r0 as our basic 'mass density')
Y i h<oo .. fr c
' t)
G o (.X '
2 > (x , t) = - 4 2 1 x.. - x..,1 d 3 x ,
,
(90)
which leads, consistent with the previous assumptions, t o a h?2� whose
maximum value is of order unity.
If we extend this program to highe r orders in Y i1 1 2 one finds that it leads
to a formal hierarchy of Poisson equations for the hf11) s of the type :
� (yi l 2hf11) ) = yµv terms + (terms known from preceding approximations) .
(9 1)
1 54 T. Damour
For the first few steps this hierarchy admits a unique solution fulfilling the
boundary conditions (89), and satisfying everywhere hfn) � 1 . However, the
terms resulting from lower approximations on the right-hand side of eq. (9 1)
quickly lead to badly behaving Poisson equations which do not admit any
solutions fulfilling the boundary conditions (89) (in the present scheme using
harmonic coordinates this problem arises first for h?J> and h?6�). The reason
for this incompatibility between the hierarchy (9 1) and the boundary
conditions (89) does not, however, mean that there is a fundamental
breakdown of the post-Newtonian approach at the level of h?J> or h?6� (called
second-post-Newtonian, or 2PN , level). Neither does it mean that the
harmonic coordinates become intrinsically 'bad' at the 2PN level . The
blame for this incompatibility lies with the boundary conditions (89).
Indeed , as was first realised clearly by Fock ( 1 959), the post-Newtonian
expansion (84)-(85) is basically a near-zone expansion of the exact hµ v (x, t),
i .e. an expansion valid only up to a radius, r, around the system which is
much smaller than a characteristic wavelength, A., of the gravitational
radiation emitted by the system : r � A.. As a consequence the asymptotic
behaviour for lxl -+ oo , t fixed , of each term of the right-hand side of eq . (84)
has, a priori, nothing to do with the real asymptotic behaviour of the exact
hµ v(x, t) (because, indeed , the expansion on the right-hand side of eq. (84)
becomes invalid when r gets large enough to reach the wave-zone r � A.). This
is clearly understood for the trivial example of the near-zone (or l/c)
expansion of the simplest radiative field :
S (t - r/c) S ( t) 1 1 1
r
- - - S ( t) +
r c 2c2
-
6c3
s·( t)r - - S ( t)r2 + · · · (92)
The program we described above (solving the post-Newtonian

expanded field equations using , often implicitly, the fall-off conditions (89)
at each step) has been developed by many authors starting with the initiator
of the PNA methods : J. Droste. See e.g. Droste ( 19 16), de Sitter ( 1 9 1 6),
Chazy ( 1928, 1 930) , Levi-Civita ( 1 937a, 1950) , Einstein, Infeld and
Hoffmann ( 1938), Papapetrou ( 195 1), Infeld and Plebanski ( 1960) (and
references therein) , Chandrasekhar ( 1965) , Chandrasekhar and Nutku
( 1969) (and other papers published in the Astrophysical Journal) , Ohta et al.
( 1973 , 1974a, 1974b). Note that those of the preceding authors who worked
at the 2PN level had to use a non-harmonic coordinate condition in order to
satisfy (89).
Other authors have tried to by-pass the difficulty of incorporating a
'good' boundary condition by replacing the hierarchy of differential systems
(9 1) with a formal iteration of integrals. One such method consists in

keeping on the left-hand side of the (expanded) field equations (9 1) a
d' Alembertian , D , instead of a Laplacian , the right-hand side being known
from preceding approximations, and then to solve
(9 3 )
by means of the formal near-zone expansion of the retarded integral of L<11l :
(() ( - ) P ap
p = 0 c p.
h c11i (x, t) = L -P-!atP IP _ 1 (L (n) ) ,
•
(94)
where for any function f (x, t)
f
I, [f] (X , t) o = f (X' , tJIX - X ' I ' d 3 '
x . (95)
The idea behind this formal procedure was to retain both the technical
simplicity (eq. (85) neglecting many terms) and the conceptual ease of the
post-Newtonian approach (hcnl being expressed as a functional of the
instantaneous state of the source, as in Newtonian theory) while still trying
to incorporate the 'real physics' (which states that in Einsteinian theory
gravity propagates with the velocity of light). This programme has been
developed by many authors, notably Peres ( 1959, 1960), Carmeli (1964,
1965), Synge ( 1969, 1970), Hogan and McCrea (1974), Anderson and
Decanio ( 1975). An improvement of the method (leaving the time
derivatives inside the integral operators in eq . (94) and 'reducing' them by
means of the equations of motion) has been proposed by Ehlers ( 1977, 1980)
and developed by Kerlick ( 1980) and Caporali ( 198 1a). However, the use of
the near-zone expanded retarded potentials (94 ) causes the appearance of
divergent integrals. This means that all these schemes break down (because
they become undefined) at the 3PN level (Kerlick , 1980), or even before in
the case of the non-improved schemes. This breakdown is a direct
consequence of the limitation of validity of the post-Newtonian expansions
to the near-zone.
Other authors, conscious of the necessity to incorporate the propagating
character of gravity (both in the near- and the wave-zone), but aware of the
simplifications brought by the PNA assumptions, attempted to devise some
mixed post-Minkowskian (see next section)-post-Newtonian approaches.
See e.g. Fock ( 1959), Persides ( 197 1a, 197 1b), Winicour ( 1983), Gurses and
Walker ( 1984), Schafer ( 1985). However, in order to really face the problem
of the near zone limitation of the post-Newtonian approach it seems
necessary to have recourse to new methods (see Sections 6. 1 1 and 6. 12).
1 56 T. D amour
On a more fundamental level there are also the problems of

understanding in general terms the transition between Einstein's and
Newton's theories (see e.g. Ehlers ( 198 1, 1984) and references .t herein) , and
of clarifying the mathematical , and numerical, validity of the post
Newtonian expansion (84) . Concerning this last point Futamase and Schutz
( 1983) attempted to prove that the first few terms of the PN expansion (84)
constituted a genuine asymptotic expansion (in the mathematical sense), but
the assumptions they used (about the existence and uniformity of the Yi --+ 0
limit, plus some other implicit assumptions) were very strong so that it is not
clear, to me, what they have actually proven . Another interesting problem is
to clarify at what order the post-Newtonian expansion fundamentally
breaks down and why (see Blanchet and Damour, 1986b, and references
therein).
6.1 1 The post-Minkowskian approximation methods

This type of method is based solely on the assumed weakness of the
gravitational field without any further assumptions concerning the
magnitude of the internal or external velocities. Using the notation of
Section 6.9 this can be written in general as :
sup(y e , Yi ) � 1 . (96)
As, from eq. (76 ), Ye < Yi one can take Y i as the main relativistic parameter of
these methods.
I favour the terminology 'Post-Minkowskian Approximation ' (PM A)
methods instead of the 'Fast Motion Approximation' , 'Weak Field
Expansion' or 'Non-Linearity Expansion' which are, or have been , also
used. The reason for this preference is that , just as the Post-Newtonian
Approximation methods were based on the idea of staying as close as
possible to the conceptual framework of Newtonian theory (separating
(absolute) time and (absolute) space, endowed with an (auxiliary) Euclidean
metric, and using instantaneous potentials), so the Post-Minkowskian
Approximation methods try to stay as close as possible to the conceptual
framework of the Minkowskian interpretation of special relativity (using an
absolute spacetime endowed with a (auxiliary) Minkowskian metric and
retarded potentials). As in the post-Newtonian case, the use of post
Minkowskian methods exposes one to the risk of forgetting the richness of
the full Einstein theory, and su rreptitiously introducing a 'neo
Minkowskian' interpretation of General Relativity.
Technically speaking , the post-Minkowskian approaches start by
writing , in a given coordinate system, the decomposition (83) of the 'real'

(Gothic) metric, gµv , into an auxiliary 'flat' metric, fµv , and a weak
'gravitational field' h µv . Then hµ v is assumed to admit a formal asymptotic
expansion when )'i � 0 of the type :
h µ v (x;.) = yih't v (x;.) + yi2 h 'ZV(x;.) + · + yi'h� v (x;.) + ·
· · · · , (97)
where the h� v (x;.) are assumed to be of order unity (or less) all over
spacetime, their first derivatives oh� v/ox;. being assumed to be of order 1/L
(or less). Then if one uses a system of units such that
c = m=L= 1, (98)
the numerical value of Newton's constant in these units is
G = Yi � 1 , (99)
and all the h�vs, and oh� vI ox;.s, as well as all the internal or external physical
quantities (masses, radii, velocities), become of numerical order unity (or
less). Therefore, analogously to the post-Newtonian case, one can summarise
the post-Minkowskian assumptions by saying that one looks for solutions
of the Einstein equations (60) in the form of formal expansions in powers of
G (and it is convenient to use this language even if one does not really use the
unit system (98)).
In the PMA method it is convenient to use a coordinate condition which
is formally Poincare invariant (so that the coordinates can still undergo a
Poincare transformation) ; this is the case with the harmonic condition (67)
which is used most often (note, however, that one could use (g gµ v) , = 0 with
w
v
any given power w, or many other conditions). Using the harmonic
condition (67) we are led to look for solutions of the system (69) of the form
(97). We can re-write eq. (69c) as a Minkowskian-type continuity equation,
(fo uµ), µ = 0, ( 100)
with a Minkowski-4-velocity uµ = uµ/( -fa.p urx uP ) 1 1 2 (satisfying fµ uµuv = - 1 )
v
and a Minkowski-proper-rest-mass density
.
fo : = r o Jg( -fa.p urx uP ) l f 2 ( 101)
Eq . (69b) will then look like a Minkowski-Euler equation containing an
explicit 'gravitational force density' starting at the order 'Yi · Finally, the field
equation (69a) leads to a forma) hierarchy of inhomogeneous wave
equations for the h� v s of the type ( 01 : 1w� a;v ) :
01 (yfh� v ) = yµ v terms + (terms known from preceding PM approximatiom
( 102)
One must now complete the partial differential equation ( 102) by a
1 58 T. Damour
prescription for choosing the 'physically correct' solutions. This is often

(implicitly) done by requiring the satisfaction of the 'Fock conditions'
( 103a)
, ..... CXl
(( ))
c + r/c = const
ohµ v 1 ohµ v
( 103b)
l
f�! t
c + r/c = const
r
0 ; +� 0 ; = 0,
where r = : ( b iixixi)1 1 2 and t = : x0/c (these conditions must be satisfied along
all past Minkowski-light-cones). Fock ( 1959) has proven that the fall-off
condition ( 103a), together with the 'no-incoming-radiation condition'
( 103b) necessarily implies that
J
y fh !'(x) = d4x' G�(x - x')S"'(x'), ( 104)
where G��tx - x') is the retarded Green function of the (Minkowski) wave
operator o1 =fµv a 2/ox µ ax v , and where sµ v denotes the right-hand side of
eq. ( 102) ('effective source of the gravitational field'). Therefore, an
alternative prescription for solving the hierarchy of differential systems ( 102)
would be to replace it by the hierarchy of iterated retarded integrals ( 104).
Note, however·, that it is not guaranteed that the 'retarded' h s defined by n
( 104) will satisfy the Fock conditions ( 103) (see Leipold and Walker, 1977).
The first step in the hierarchy ( 104) was . introduced and explicitly
worked out by Einstein (1916b); this is the famous 'linearized
J
approximation' to general relativity :
1 6nG �
Yi h1:V (x) = d 4x' G�ft> (x - x')Tµ v (x'), (105)
c4
where, following the notation of eqs. ( 100}-( 10 1), 'f'µ v denotes a Minkowski
like stress-energy tensor of the matter.
For a long time the post-Minkowskian approximation stayed dormant,
while the post-Newtonian one was developed. It was revived in 1956 and
then began to develop in turn : see Bertotti ( 1 956), Havas ( 1957), Kerr
(1959), Bertotti and Plebanski (1960) (first attempt to explicitly tackle the
non-linear approximation), Havas and Goldberg ( 1 962), Kuhnel ( 1963,
1964), Stephani (1964), Goenner ( 1 970), and Bennewitz and Westpfahl
( 197 1). A new approach to the post-linear formalism, based on an attempt to
use retarded Green's functions in curved space-time (however, the final
formulas use only the fla t Green function) , was devised by Thorne and
Kovacs (1975) and Crowley and Thorne (1977). Recently several aut hors
have succeeded in computing explicit expressions for the equations of

motion and gravitational field in the second post-Minkowskian
approximation : Westpfahl and Goller ( 1979), Westpfahl and Hoyler ( 1980),
Bel , Damour, Deruelle, Ibanez and Martin ( 198 1), Deruelle ( 1982),
Westpfahl ( 1985) ; and even partly for the third post-Minkowskian
approximation : Damour (1982, 1983a, 1983c). More recently a
combination of the post-Minkowskian approach with multipole expansions
has allowed one to control the post-Minkowskian expansion (97) at all
orders of approximation , see Blanchet and Damour (1984a, 1986a) and
references therein. This result strengthens the hope that the post
Minkowskian approach is a 'good' approximation method which, contrary
to the post-Newtonian one, can be unambiguously iterated to all orders, and
which yields a useful sequence of approximate metrics all over spacetime
(the fact that this sequence might not be convergent but only asymptotic
(Christodoulou and Schmidt, 1979) is not a problem for its use in physics,
and the weakly unsatisfactory behaviour of P M expansions in the very far
wave-zone is easy to correct, see next section).
It should also be noted that Schafer ( 1986) has recently introduced a non
linear expansion which is based on the Arnowitt-Deser-Misner
hamiltonian approach (Arnowitt et al. , 1962) and which seems to combine
some of the advantages of the post-Newtonian and post-Minkowskian
approaches without having the defects of the post-Newtonian one (near
zone-validity only).
Finally, one can remark that the post-Minkowskian approach can be
thought of as a 'meta-approximation-theory' for the post-Newtonian one.
Indeed, when the assumptions (80) are satisfied one can use units such that
sup (vi nt , vext ) = m = L = 1 , ( 106)
for which the numerical value of the velocity of light is very large, the
numerical value of Newton's constant is ::5 1 , and
G
Yi = 1 � 1 . ( 107)
c
We can then try to re-expand each term of the PM expansion,
h µv (x ;.) = L ��1 h�v (x, t, �) ,

n ;.. I C C
( 108)
in a second (post-Newtonian or near-zone) asymptotic expansion when

l/c O. It is precisely from this point of view (descending from PMA to
--+
PNA) that one 'understands why' the PN expansions have a restricted

1 60 T. Damour
spatial range of validity (r � ,\). Using this point of view one can similarly
investigate the domain of validity of the null-cone-improved post
N ewtonian approaches of Persides ( 197 la, b) and Winicour ( 19 83). These
methods essentially consist of looking for expansions of the type
hµv (x ;. ) = L
m ;;;i: 2
: k�v (x , u),
C
( 109)
where u is some retarded time ( � t - r/c) which is kept fixed in the limit
process 1 /c -+ 0. Now the expansion ( 109), if it exists, must 'descend' from the
PM expansion ( 108), re-written in terms of (x , u), and re-expanded for
1/c 0. Then explicit calculations at the 2PM level (in general relativity,
-+
and in some scalar field models with ,\</J" non-linearities) show that these
improved approaches have a range of spatial validity which is still restricted
to the near-zone (r � ,\) (only the linearised approximation sees its range of
validity extended to the wave-zone). In particular, this can be used to show
that the 'logarithmic violation' of peeling found by Isaacson, Welling and
Winicour ( 19 84) is only a near-zone effect (behaviour when r � R but r � A)
and says nothing about the 'real ' asymptotic behaviour of the gravitational
field at future null infinity.
6.12 Singular perturbation methods

The 'breakdown' of the post-Newtonian expansion (84)-(85) (formal
expansion in powers of the small parameter 1 /c , see eq . (87)) outside of the
near-zone (see eq. ( 92)) is an example of what goes under the general name of
'singular perturbations' : i .e. an asymptotic expansion, when a 'small'
parameter E tends to zero, of some field quantity,
( 1 10)
n
(where bn + 1 (c)/bn(E) -+ 0 when E -+ 0), such that some of the coefficients fn(x)
are not bounded on the domain of variation of the configuration-space
variables x. Problems of this type are quite common in physics, particularly
in fluid mechanics , and several methods have been developed to cope with
them (see e.g. Lagerstrom et al. , 1967; van Dyke, 1975). There are essentially
two classes of methods. In one class one looks for a generalised asymptotic
expansion,
( 1 1 1)
II
admitting generalised coefficients gn(x ; E) which are bounded all over the
configuration space (this class includes the use of a-dependent changes of

configuration space variables gn = gn( Y) with y = y(x; a)). In another class one
uses, simultaneou sly, several ordinary (Poincare-type) asymptotic
expansions corresponding to the use of several different configuration-space
variables related via a-dependent transformations, for instance
f- Ln bn(a)fn(x), ( 1 12a)
( 1 12b)
n
where X = X(x, a). The subtlety of the latter approach lies in the fact that
when a ---+ 0 the a-dependent domains of the configuration-space, where the
expansions ( 1 12a, b) are a priori valid, are not required to overlap (and in
general they do not).
One can say that , as seen from the post-Newtonian point of view, the
post-Minkowskian expansions constitute a first class 'cure' for the wave
zone breakdown of PN expansions (see eq. (108)). Now the usual post
Minkowskian expansion itself exhibits a kind of (weak) breakdown in the
'exponentially far-wave zone'
{ }
r ;;:: A. exp
c 2 A.
4nGm
, ( 1 13)
because of the appearance of log r/r terms, when r --+ oo , in h1r. However, as
argued already by Fock ( 1959), Isaacson and Winicour ( 1 968), and more
recently by Anderson ( 1980) and others, this can easily be cured by a first
class method based on a (G/c 2 )-dependent change of coordinates (for a
complete proof that this cu re works at all orders in G , see e.g. Blanchet
( 1987)).
The singular-perturbation methods of the second class have been found
useful in several problems of relativistic gravity. They have been used, in an
intuitive way, by Fock ( 1 959, section 87) and Manasse ( 1963); and were
introduced in a more formal way in the late sixties by Burke ( 197 1) and
Thorne ( 1 969) to tackle the wave-zone breakdown of PN expansions.
Recently they have been used by several authors : e.g. Demianski and
Grischchuk ( 1974), D'Eath ( 1975a, b), Kates (1980a, b), Anderson, Kates,
Kegeles and Madonna ( 1982), Damour ( 1983a), Anderson and Madonna
(1983); Blanchet and Damour (1984b), Thorne and Hartle ( 1985), Futamase
( 1986). A general discussion of the formal way in which these methods fit
into the framework of General Relativity has been attempted (Kates, 1 98 1 ;
162 T. Damour
Damour, 1987). However, in my opinion, there are still fundamental issues

concerning this second class of methods which have not been clarified.
Indeed , the basic paradox of these methods is that the domains of a priori
validity of, for instance, the two asymptotic expansions ( 1 12a) and ( 1 12b)
become completely disjoint when e 0, but still they are supposed to
__.
describe fully the physical problem at hand so that they must somehow fit
together. Now the literature which deals with precise rules for the 'matching
of asymptotic expansions' is full of ambiguities and controversies. In the
context of General Relativity some of these ambiguities have been pointed
out recently (Blanchet and Damour, 1984b; Damour, 1987). In particul ar, a
re-examination of the problem initially studied by Burke ( 197 1) has shown
the necessity to take non-linear effects into account with great care (Blanchet
and Damour, 1984b, and references therein). This is why , when using this
type of method in the problem of the motion of condensed bodies, care has
been taken to control in sufficient detail the structure of the non-linear
corrections so as to be able to extract minimal, but reliable, information
from the matching of the two expansions (Damour, 1983a, sections 5 and 6).
6.1 3 The external and the internal problems of motion (Einsteinian

case)
Returning to the problem of motion of N widely separated bodies in general
relativity, it is clear from the study of the corresponding Newtonian case
(Section 6.3) that we should, at some point, try to split the problem into two
sub-problems, one concerning the motion of each body as a whole ('external
problem') and the other concerning the intrinsic evolution of each body
('internal problem'). However, the fact that there are more potentially small
Ye , Yi, Pe , f3J
parameters in the relativistic case (ar:, than in the Newtonian case
(ar:) means that there are several different ways of doing this, which
correspond , roughly speaking, to considering the non-relativistic and
relativistic parameters in a different order.
The most commonly used way, implicitly influenced by the 'neo
Newtonian' (or the 'neo-Minkowskian') reinterpretation of general
relativity, first considers the relativistic parameters and uses them to 'force'
Einstein's theory to fit the Procrustean beds of Newtonian theory or of special
relativity. Having thus recast the problem in the framework of an auxiliary
absolute spacetime (£ 3 x 1R or M 4) the further consideration of the
separation parameter cir: leads one 'natu rally' to first seek a 'good'
generalisation of the Newtonian centre of mass (7), and then to use this
generalised centre of mass to formulate the external problem (the internal

problem being supposed to follow without requiring special attention).
6. 1 3 . l The centre of mass problem
Several different approaches to the definition of a 'good' centre for each
body have been pursued. Some authors start by imposing some 'symmetry'
(spherical or spheroidal) on each of the bodies and then consider the 'centre
of symmetry' . See e.g. De Sitter ( 19 16), Levi-Civita ( 1937-50), Papapetrou
( 195 1), Misner, Thorne and Wheeler (1973), Hogan and McCrea (1974),
Caporali ( 198 lb) . The danger with this approach is that the 'symmetry' is
defined with respect to the auxiliary absolute space. But this means, in
'reality', that it is a symmetry with respect to the arbitrary global coordinate
system . Therefore , it is just a fictitious 'coordinate symmetry' which has, a
priori, nothing to do with the real physical symmetry of each object. On the
contrary it is clear, because of the anisotropic 'Lorentz contraction' of
moving bodies, that to impose a coordinate symmetry on moving objects
means that, 'in reality', they are non-spherical (even when real tidal or spin
effects are negligible) �o that this requirement gives them in fact an unwanted
'ellipticity' of order p;, and thereby an intrinsic quadrupole moment of
order mL2 {3; which gives an (erroneous) contribution to the equations of
motion of the bodies of the order:
i i
Gm 2
b Force -
RZ a f3e . ( 1 14)
Note that this contribution is of order 1/c 2 ( lPN order) and can become
important if a is not very small or if one works to a higher post-Newtonian
order. Note also that some authors are free of the preceding criticism, either
because they take into account the Lorentz contraction explicitly (Lorentz
and Droste, 19 17), or because they endow the bodies with internal stresses
suitable to give them the required coordinate symmetry (Hogan and
McCrea , 1974; McCrea and O'Brien, 1978).
Other approaches to the definition of a 'gooa' centre of mass seek to
imitate the Newtonian definition (6)-(7). The problem is thus reduced to
finding a 'good' relativistic analogue, say p, of the Newtonian mass density
such that the centre of mass reads :
( 1 15)
with
( 1 16)
1 64 T. Damour
One of the initiators of this approach is Fock ( 1 959). He begins by using

the coordinate rest-mass density, eq. ( 8 2 ) (denoted p by him) :
p 1 = = r* , ( 1 17)
but after having obtained the external equations of motion he introduces an
'effective mass' and an effective centre of mass (Fock, 1 959, eqs. (78 . 1 1),
(78 . 12); see also Bromberg, 1972, chapter 7) which correspond to modifying
eq. ( 1 17) by taking into account both internal energy and self-gravitational
energy relativistic corrections. More recently, it has become customary to
start directly with such improved effective definitions of the 'mass' and the
'centre of mass' with the help of the following effective relativistic mass
density for the ath body (Will , 1974; Contopoulos and Spyrou , 1976;
Spyrou , 1977) :
( 1 18)
using, respectively, the definitions ( 1 7)-( 18), ( 1 5) and (66) for the internal
velocity, self-gravitational potential and specific internal energy density .
Other authors use a purely formal approach to the centre-of-mass
problem in the sense that they introduce from the start a distributional
energy-momentum tensor with support located on some world-line (see e.g.
Infeld and Plebanski ( 1960), Ohta et al. ( 1973), Havas ( 1 979) and references
quoted in these works). However, the crucial issue of justifying the (a priori
meaningless) use of such a distributional energy-momentum tensor on the
right-hand side of the non-linear Einstein field equations has , strangely
enough, rarely been seriously considered . Until recently the only
'j ustification' for these approaches was a posteriori, in the sense that they
gave final results in agreement with other 'orthodox' methods, at least at the
lowest relativistic orders ( lPN or lPM). And still these methods seem a
priori attractive because they simplify the calculations considerably and allow
one either to quickly recover results known from complicated orthodox
approaches or make explicit computations at the higher relativistic orders
(2PN (Ohta et al., 1973 , 1974); 2PM (Bel et al. , 198 1 ; Westpfahl , 1985)). A
preliminary study of the inner consistency of the use of 'delta functions' in
General Relativity has been initiated by Bel et al. ( 198 1). But, even if one
stays at the purely formal level , there seem to be obstacles to the use of 'delta
functions' if one works at the third post-Minkowskian level . However,
because of the efficiency of these approaches an effort has recently been
made to bridge the gap between orthodox and distributional approaches.
For the moment there exists no result which is valid at all orders, but it has
been possible to prove (Damour, 1980, 1983a) that the calculations within a
sound orthodox approach (combining a minimal matching of asymptotic

expansions and an Einstein-Infeld-Hoffmann-Kerr-type approach) could
be very much simplified by a mathematical trick (complex analytic
continuation) which makes the calculations resemble formal 'delta function'
calculations (see Section 6. 14). However, I wish to emphasise that this
resemblance is used only to simplify the explicit computations within a
theoretically sound approach and does not mean that one is using a
powerful , but otherwise arbitrary 'regularisation procedure' (there seems to
have been some misunderstanding of this point by some authors, e.g.
Futamase ( 1983), Schutz ( 1985), Grishchuk and Kopejkin (1986)).
One should also mention the work of Dixon ( 1979), Ehlers and Rudolph
( 1977) and Schattner ( 1978) on a covariant approach to the definition of a
centre-of-mass world-line, although it is not clear how to relate this
approach to the other, less covariant but more explicit, approaches.
Finally, a very different way of attacking the centre-of-mass problem
proceeds the other way around. Instead of, first, defining the centre of mass
as a tool to separate the problem of motion into external and internal parts,
there are approaches which initially separate the external problem of
motion , and then pin-point , within their external scheme, a convenient
'central point' . The initiators of this type of approach were Einstein, Infeld
and Hoffmann ( 1 938) (see also Einstein and Infeld , 1940, 1949). Modern
disciples of the approach use the 'method of matched asymptotic
expansions' (see e.g . D'Eath, 1975b; Damour, 1983a; and for a somewhat
different approach Thome and Hartle, 1985). In these approaches the
'central point' appears more as a 'centre of (gravitational) field' than as a
'centre of mass'.
After choosing some definition of the centre of mass of each body one can
then proceed to derive its equations of motion and to discuss the internal
problem.
6 . 1 3 .2 Results in the external problem
It is outside the scope of this article to give a detailed account of all the
results which have been obtained in the Einsteinian external problem of
motion. Some of the recent results in the two-body problem will be discussed
below (Section 6 . 1 5). We shall content ourselves here with giving a bird 's
eye view of the work done and some references.
First let us recall that the general structure of the external equations of
motion will be :
d 2z
A y ( 1 19a)
dt 2 = G (r1., e , Yi , f3e , f3J
166 T. Damour
cx
Now the parameter must necessarily be small in all situations where one
wants to talk about an external problem in contrast to an internal one.
Moreover, as Ye ;5cx
(eq. (78)) there are certainly at least two small
parameters among the five appearing in eq. ( 1 1 9a). Therefore we can expect
that, in the most general case, it should be possible to write the following
expanded form of the external equations of motion :
= G m Amn(Yi ,/3e , f3Jcxmy�.

d 2z
L ( 1 19b)
dt 2 , ;;i. O
n
In order to simplify our presentation we shall introduce only powers of the

small parameters, but the presence of more complicated scale functions for
the expansion (such as cxm(log cxr· ,. . . ) can easily be coped with. This kind of
expansion ( 1 19b) does not require that the internal gravity
velocities, Pe , /3i,
be small. It is this kind of expansion which appears in some
Yi, or the
post-Minkowskian approximation schemes for the external gravitational

field, completed by some matching to the possibly strong internal
sources). A term of the type

coupling constant G, G Amncxmy:,
gravitational field (sometimes formally modelled by the use of distributional
which contains (n + 1) powers of the
is said to belong to the (n + 1)-PM approximation
level. The usual post-Minkowskian schemes (requiring everywhere-weak
gravitational fields without conditions on the velocities) would mean that
one is further expanding eq. ( 1 19b) in powers of Yi :
d 2z
dt 2 =GL
m , n , p ;;l= O
Bm11p(/3e , f3Jcxmy:yf. ( 1 19c)
Such schemes have never been developed in detail . On the other hand , the
usual post-Newtonian schemes (requiring everywhere-weak gravitational
fields and small velocities) lead to a fully expanded form :
d2z
dt 2
=G m, n, p", q, r ;;l= O CmnpqrCXmYenYip/3eq/3ir
Li ( 1 19d) •
A term of the type GCmnpq rcxmy:yf/3�/3[, which contains (2n + 2p + q + r)

powers of l/c, is said to belong to the (n + p + i(q + r))-PN approximation
level .
Explicit results of the minimally expanded form ( l 1 9b) have been
obtained at the lPM level by several authors, starting with the pioneering
works of Bertotti ( 1956) and Havas ( 19 57), see references in Havas ( 1979),
Damour ( 1983a) and Westpfahl ( 1985). For explicit results at the 2PM level
°
see WestpfahI and Goller ( 1979), Bel et al. ( 198 1) and Westpfahl ( 1985). The
3PM level has been partially tackled (Damou r, 1982, 1983c) keeping only
the terms necessary to derive the 2 . 5PN equations of motion (of the form
( 1 19d) without Yi-expansion). The lPM equations of motion of spinning
bodies have been studied by Ibanez and Martin ( 1982).
Explicit results of the fully expanded form ( 1 19d) have been obtained at
the lPN level by many authors starting with the pioneering work of Lorentz
and Droste ( 19 17). Note that the latter work is the first one to give the
correct l PN equations of motion, the earlier results of Droste and De Sitter
and the later results of Chazy and Levi-Civita are incorrect. For references
to more recent works see e.g. Brumberg ( 1972), Ehlers ( 1977 , 1984),
Caporali ( 198 1b), Damou r ( 1983a), Grischchuk and Kopejkin ( 1986) and a
long-promised (but not yet published) review article by P. Havas. Special
mention should be given to the problem of the motion of spinning bodies
which involves several subtle issues (centre of mass, ordinary versus
acceleration-dependent Lagrangian). This problem has been reviewed by
Barker and O'Connell ( 1 979). Further references to the works of Michalska,
Gabosz, Lanzano and B romberg will be found in Bromberg (1972). Recent
works are due to Damour ( 1978, 1982), Dallas ( 1977), Bel and Martin
( 198 1), Ibanez and Martin ( 1982) and Thorne and Hartle ( 1985).
Concerning the issue of the presence of accelerations in the Lagrangian for
spinning bodies and of their avoidance by use of a suitable non-covariant
spin-condition see e .g. Damour ( 1982).
Explicit results at the 2PN level have been obtained by Ohta et al. ( 1974b)
for N point-masses, by Damour and Deroelle ( 198 la) and Damour ( 1 982,
1983a) for two condensed o bjects, by Schafer ( 1985) for N point-masses, and
by Kopejkin ( 1 98 5) and Grishchuk and Kopejkin ( 1986) for two weakly self
gravitating fluid balls. The 2PN equations of motion can be derived from a
Hamiltonian as first shown by the work of Ohta, Okamura, Kimura and
Hiida ( 1974b). However, there are several subtle issues connected with this
Hamiltonian and with its Legendre-transform associated Lagrangian. On
the one hand , one term in the potential part of this Lagrangian, for the two
particle case, has been wrongly evaluated in Ohta et al. ( 1974a), as was first
remarked in Damour ( 1982). The correct evaluation has been given in
Damour ( 1983a) and Damour and Schafer ( 1985). More important is the
issue of the influence of the choice of the coordinate system on the
Lagrangian. Indeed, it has been shown by Damour and Deruelle ( 198 lb)
and Damour ( 1982) that the 2PN motion of two condensed bodies in
harmonic coordinates could be deduced only from a generalised Lagrangian
function involving not only positions and velocities bu t also
accelerations. This issue has been recently clarified by Damour and Schafer
168 T. Damour
( 1985), following a remark of Schafer (1984) that velocity-dependent

coordinate transformations introduce an acceleration dependence in the
Lagrangian. In particular it has been explicitly checked that the ordinary
Lagrangian of Ohta et al. ( 1974b), corresponding to a non-harmonic
coordinate system, is indeed equivalent (in the two-body case) to the
generalised Lagrangian of Damour ( 1982), corresponding to a harmonic
coordinate system. On the other hand, the ordinary Lagrangian of Ohta et
al. ( 1974a) is incorrect ; it should d �pend on accelerations.
Explicit results at the 2 .5PN level have been obtained by Damour ( 1982,
1983a), Schafer ( 1983 , 1985), Kopejkin ( 1985) and Grishchuk and Kopejkin
( 1986). These will be discussed in Section 6 . 1 5 .
An important issue which distinguishes the fully expanded post
Newtonian equations of motion ( 1 19d) (as well as the intermediate post
Minkowskian expansion ( 1 19c)) from the minimally expanded eqs. ( 1 19b) is
the explicit appearance of the internal field-strength parameter 'Y i . This
means, therefore, that the expansions ( 1 19c) or ( 1 19d) can be applied a priori
only to weakly self-gravitating bodies. This limitation is very serious
because modern astrophysics often has to deal with the motion of strongly
self-gravitating bodies (also called condensed bodies, like neutron stars or
black holes). However, several authors (McCrea, Grishchuk and Kopejkin,
. . . ) have argued that it was justified to use the post-Newtonian method even
for strongly self-gravitating bodies because explicit calculations showed that
the way the parameter 'Yi entered the eqs. ( 1 19d) was such that it could be
incorporated into some 'effective' mass m = m 0 + yim 1 + · · · (this 'effacement'
property will be discussed in more detail below (Section 6. 14)). After this
'renormalisation' of the mass the resulting equations of motion contain the
external relativistic parameter ye but no longer contain the internal one 'Yi , so
that they seem to apply as well to weakly as to strongly self-gravitating
objects. However, I wish to point out that , although this argument is
certainly correct (except for black holes) if one proves the 'effacement' of the
yis to all orders of approximation , it is definitely insufficient if one proves the
effacement of the yis only up to some finite order. More precisely, we are
going to show the surprising result that the usual way of truncating and then
renormalising the PN expansion ( 1 19d) is valid only at the first post
Newtonian approximation and is already inconsistent at the 2PN level !
Indeed, in all explicit post-Newtonian calculations of the external
equations of motion, working say at the nPN order, one is discarding not
only all the terms coming from the higher PN levels but also many terms
depending on the separation parameter, a , so that the uncontrolled relative
error term in an explicit nPN equation of motion will be

e <; ex 5 + ex 2 p; + y j + 1 , ( 120a)
where the first term comes from the tidally induced quadrupole force
between deformable bodies (see Section 6. 14), the second term is the effect
discussed in eq . ( 1 14) which should be included in a consistent calculation
using coordinate-spherical bodies (as often assumed), and the third term is
expected to be the dominant term coming from the (n + l)PN
approximation (because p; - Y e � yJ. We are here neglecting the fact that at
higher PN approximations the error terms seem no longer to be simple
powers of the PN parameters but to include some logarithms. Now the point
is that ex and Yi are not independent because Yi = Gm/c 2L = (Gm/c 2 R) x
(L/ R) - 1 = Ye ex - 1 , so that for a given external relativistic parameter, Ye "' p;,
the internal gravitational field strength, Yi , increases when the bodies
become more separated (note that this means that the expansions ( 1 19c) and
( 1 19d) were not real power expansions with respect to several independent
parameters but were defined, for instance, by the re-expansion of the
original coefficients Am (Yi ) as Yi � 0). For simplicity's sake let us consider
n
only the first and the third term on the right-hand side of ( 120a) because they
are there independently of the wrong use of coordinate symmetry (the taking
into account of the second term strengthens the conclusions below without
changing them qualitatively). Then the relative error, e, will certainly exceed
a minimum value reached when ex 5 "' Yi + 1 = y� + 1 ex -" - i , i .e. for an optimal
ex - y�n + 1)/(n + 6 > . Hence
(120b)
with
n+ l
p= S . ( 120c)
n+6
Now if n = 1 ( l PN approximation), p = 17° > 1 , so that it is consistent to
retain the lPN corrections in the external equations of motion ( - p; + Y e )
and to discard all the other uncontrolled terms hidden in e . 'Consistency'
means here that there exists a limit process, in which Y e , Y i , and ex tend to
zero, for which e becomes asymptotically negligible with respect to the terms
explicitly retained . On the other hand , if n . 2 (2PN level), p = 185 < 2, so that
for any limit process the last terms retained in the external equations of
motion ( - (p; + Y e ) 2 ) are necessarily negligible compared with the
uncontrolled error terms contained in e. And this situation gets worse as n
increases. This conclusion does not mean that the final 2PN or 2 .5PN results
obtained by several authors (explicitly dealing with extended fluid bodies)
1 70 T. Damour
are incorrect, but it means that they have not yet been justified within the
PN-fluid schemes. More work is needed to control the effect of the higher
order self-field effects.
6 . 1 3 .3 The internal problem

In Newtonian gravity the internal problem means : 'to determine the motion
of each body around its centre of mass' . Technically this means (Section 6.3)
introducing the relative coordinates, y� , expressing the position of each
material point of the ath body with respect to the chosen centre of mass,
z�(t), so that (see eq . (17))
x i = z�(t) + y� . ( 12 1)
Many authors, working in the post-Newtonian framework , have, more or
less implicitly, assumed that the mathematical formula ( 12 1) was sufficient
to express the passing to a 'centre-of-mass frame' in Einsteinian gravity. In
principle this view is admissible because the coordinate systems are
arbitrary in General Relativity. This is the view which is often taken in
relativistic celestial mechanics to express, for instance, the motion of the
Moon , or of an artificial satellite, with respect to the Earth (see e.g .
Bromberg, 1972 ; Lestrade et al. , 1982; Martin et al. , 1985). H owever, in
order to be able to use this viewpoint consistently one must also write the
equations for the internal problem of each body using the spacetime
coordinates (y�, t = x 0/c) and solve them with the adequate precision . For a
long time this internal problem was neglected because the accuracy of the
measurements allowed one to use the results of the Newtonian internal
problem, the relativistic 'corrections' being taken into account only in the
external problem (and considered only through their secular effects) .
However, the great increase in accuracy of the measurements now available
(concerning both the dynamics of natural or artificial bodies in the solar
system, and the dynamics of binary pulsars) obliges us to tackle seriously the
problem of a fully consistent relativistic description of the internal problem .
Then if one tries to really use the coordinate system (y� , t) defined by eq .
( 12 1) (with z� = z� being one's preferred relativistic centre of mass) this will
lead to an internal problem which will have a structure of the type :
d
i
0 = i + Io(Xj ) + p;11(Xi , Xe) + yel"(Xi , Xe) + . . . + c.t 3 1 3 {X i , Xe) + . .
·,
( 122)
to be compared with the Newtonian equation (55b). In the Newtonian case
the influence of the external structure on the internal problem was effaced up
to the order cx 3 . But here we see that there already i s a strong influence of the
external motion on the internal one of order p; or 'Y e · The presence of the p;
coupling is readily understood if one remembers that with the
transformation ( 12 1) one is using the external coordinate time, t, as internal
time. Then this means that each body will suffer an apparent 'Lorentz
contraction' in the direction of external motion by a factor
J{ l - (vext/c) 2 } � 1 -i(vext/c) 2 . Similarly, the Ye-coupling in eq. ( 122) can be
related to what I will call the apparent 'Einstein-contraction', i .e. the factor
linking the coordinate length to the proper length, as was first discussed by
Einstein ( 19 16a). In harmonic coordinates this apparent 'Einstein
contraction' of a body embedded in an external gravitational potential U (e)a
is isotropic and is given by a factor � 1 - U (e)a/c 2 . The Lorentz contraction
effects mean that (when real tidal effects are negligible or subtracted) a point
on the Earth's surface suffers, in the global post-Newtonian coordinate
system, an anisotropic (coordinate) displacement of maximal amplitude (in
the direction of the velocity of the Earth vE9)
�(":)2R $- 3 cm .
Such a displacement is not negligible in some of the very precise modern

measurements.
As for the 'Einstein-contraction' effects they are less dangerous because
they are isotropic. Because of the small eccentricity of the Earth's motion
they produce an apparent radial contraction of Earth of mean value ,.., 7 cm
with a yearly variable part of millimetric amplitude. It is, however,
historically interesting that it is essentially neglect of this Einstein
contraction which caused De Sitter ( 19 16) to make a mistake in his
derivation of the equations of motion of N bodies at the fir�t post
Newtonian approximation. On the other hand, Lorentz and Droste ( 19 17)
carefully take into account both the 'Lorentz' and the 'Einstein' contraction,
correctly pointing out that, when real tidal effects are negligible, intrinsically
spherical bodies will appear, in the coordinate system, as ellipsoids of
coordinate volume v � V0 ( 1 -i(vext/c) 2 - 3 U(el/c 2 ). See Martin et al. ( 1985),
and H ellings ( 1986) for heuristic discussions of these effects in the modem
context of global post-Newtonian descriptions of laser tracking of satellites
and VLBI.
All the preceding subtleties (see also ( 1 14)) indicate that the usual global
post-Newtonian way of approaching the internal problem may be, in the
long term , more trouble than help. At this point it is worth looking back
172 T. Damour
carefully at eq. ( 12 1) which, beyond its simple mathematical structure, hides

a non-trivial physical phenomenon, namely an accelerated frame of
reference with respect to which the external gravitational effects are strongly
effaced, leaving only small tidal effects (Section 6.5). Now, if we were
considering the internal problem of a negligibly self-gravitating body, the
relativistic analogues of such an 'external-gravitational-field-effacing' frame
would be the well-known 'locally inertial frames' which can be explicitly
constructed by means of Fermi coordinates based on the centre-of-mass
world-line (see e.g. Synge, 1960; Misner, Thorne and Wheeler, 1973). If we
consider, however, a self-gravitating body 'a ' , then there exists, at present,
no general definition of such an 'external-gravitational-field-effacing' frame,
and in fact no proof that such a definition is possible. However, it has been
remarked by Bertotti ( 1954) that , at the first post-Newtonian
approximation, the external motion of a self-gravitating body , 'a ' , could be
interpreted as the external motion of a test body (non-self-gravitating) in
some effective external gravitational field, say g:�a (this property is related to
the effacement of internal structu re in the external problem , see next
section). This result suggests that a good definition of an 'extemal
gravitational-field-effacing' frame consists of constructing a 'locally inertial
frame' in the effective external field g:�a based on the centre-of-mass world
line of the ath body. At the heuristic level such a construction has been more
or less explicitly assumed by many authors (see e.g. Will , 1974; Misner et al.,
1973 ; Will and Eardley, 1977 ; Mashhoon , 1984, 198 5 ; Thome and Hartle,
1985). The study of the motion of strongly self-gravitating bodies (neutron
stars or black holes) has also made use of such 'external-field-effacing
frames' (Y i , T) linked with the external coordinate system xµ by
transformation formulas of the type :
(123)
instead of the Newtonian-type formula (121) (see e.g. D'Eath ( 1 975b), eq .
(2. 10), Damour ( 1983a), section 6, eq . (7)). Recently Ashby and Bertotti
(1984, 1986) have explicitly constructed, within the first post-Newtonian
approximation, the Fermi coordinate transformation (of the type ( 123))
based on the effective external metric g:�a for the motion of a self-gravitating
body, and have verified that it has the good 'external-field-effacing'
properties. This explicit result supports one's expectation that a good
internal frame exists in the general case, but much work remains to be done
to prove this expectation. The 'internal frame problem' has also recently
become important in the relativistic celestial mechanics of the solar system
(see the contributions of Bertotti, of Boucher, of Fukushima, Fujimoto ,

Kinoshita and Aoki, and of others in Kovalevsky and Brumberg ( 1986)).
6. 14 The effacement of internal structure in the external problem

(Einsteinian case)
As recalled in Sections 6.4-6.6, several independent circumstances (Action
and Reaction Principle, Equivalence Principle, properties of multipole
expansions) imply that the influence of the internal structure on the external
motion is strongly reduced by a factor a 4 from a priori order of magnitude
expectations. It was remarked by M . Brillouin ( 1922) that the arguments by
which one can 'efface' the internal structure in Newton's theory cannot be
immediately translated into Einstein's theory (especially because of the
strong-non-linearity of the latter). This problem has been clearly formulated
by Levi-Ci vita ( 1937a, 1950). He proved by detailed arguments within
Einstein's theory that this 'effacement' property was still valid at the first
post-Newtonian approximation, in the sense that : all large direct self-action
effects cancel or contribute terms in the equations of motion which can be
renormalised away, so that the final equations of motion can be written in
terms of only some 'centres of mass' and some 'effective masses'. Actually
these results had been previously obtained, though only implicitly, in the
first correct work on " lPN equations of motion : Lorentz and Droste ( 19 17).
On the other hand, many works on the equations of motion (De Sitter,
19 16; Chazy, 1930; Eddington and Clark, 1938) had assumed the effacement
property without proof. Later, Fock ( 1959) remarked that the effective mass
of each body, at lPN order, can be written as
m. = t ,.[ 1 + c12 H (w�) 2 - �

d3x ui·� +rlo n ( 124)
i.e. as the sum of its total rest-mass energy, of its internal kinetic energy, of its
self-gravitational energy and of its internal elastic energy.
Recently the extension of the property of effacement beyond the first-post
Newtonian ( lPN) approximation has been studied further. On the one
hand, some authors, staying within the post-Newtonian framework (i.e. for
weakly self-gravitating, slowly moving bodies) have shown that it still holds
at the 2!th post-Newtonian approximation (Grishchuk and Kopejkin,
1986 ; and Kopejkin, 1985, improving on previous partial results of Rudolph
and Borner; and Breuer and Rudolph, 1982). They have shown (in the
absence of internal motion, and for 'spherically symmetric' bodies) that the
effective mass is equal , up to order c - 6 , to the 'Tolman mass' of each body.
174 T. Damour
On the other hand, other authors (D'Eath, 1975b ; Damour, 1983a ; Thorne
and Hartle, 1985) have shown that the effacement property still holds for
strongly self-gravitating bodies. In particular, it has been proven that the
equations of motion of non-rotating 'elastic' condensed bodies (neutron
stars or black holes), which would be spherically symmetric if isolated, can
be written in terms of only some 'centres of field' and some 'Schwarzschild
masses' (Damour, 1983a) up to the appearance of very small internal
structure dependent terms of relative order y; in the equations of motion (i.e.
tenth post-Newtonian order in the usual counting !). The intuitive reason for
this extremely good effacement is that each 'elastic' body is supposed to be in
a stable spherically symmetric state when isolated so that when it is not
isolated it becomes distorted by 'tidal forces' GmL/R3 , thereby acquiring a
()
small (structure dependent) ellipticity
8 _ tidal gravity GmL/R3 L 3
= = er: 3 ·
self gravity Gm/L2 R
Then the corresponding tidally induced quadrupole moments of each body,
amL2 , introduce some (structure dependent) interbody forces - am2L2/R4
which are therefore aL2/R2 - (L/R)5 = rx5 smaller than the main Newtonian
forces. Now, �y definition, condensed bodies are such that L - Gm/c2 , so
that
L Gm
er: = ,...., =y ,
R c2R e
0
hence the very small relative correction cr: 5 - y� = (Gm/c2 R)5 = O { l/c1 ) =
lOPN order.
It has been noted that the property of effacement of the internal structure
in the external motion of condensed bodies seemed to be a very peculiar
property of Einstein's theory and that in other relativistic theories of gravity
the equations of motion of condensed bodies already contain, at the l PN
level (0( 1/c2) relative corrections), some internal-structure-dependent
parameters (see Will, 198 1 , and references therein). It is therefore worth
trying to understand in a non-complex way what are the elements of
structure of general relativity which allow the effacement of internal
structure. In the proofs quoted above these elements of structure are
somewhat lost in the midst of long calculations or complex arguments.
However, it is possible to present a proof of the effacement property within
Newtonian theory which is equivalent but different from the usual
straightforward one presented in Section 6.5 and which shows up some of
the basic elements of structure which allow one to prove the effacement
property within Einstein's theory. I shall therefore spell out some details of
this proof which can serve as a nice model problem for understanding some
of the 'inner wheels' of general relativity.
The basic trick is to rewrite the Euler. equations (3) in the form of a
conservation law (using the continuity equation (2) and Poisson's law (4a)) :
a . a . . a .. a ..
- (p v1) + -. (p v 1 v' ) + -. T 11 + -. t11 = 0, ( 125)
at ax1 ax1 ax1
where
Tii + p (J ii '
=
( 126)
[ ]
is the material stress tensor, and where
1 a u a u � tJ . . a u a u
tii = + -- - ( 127)
4nG ax i axi 2 IJ axk axk
is a quadratic form in the gravitational field, which is the gravitational
analogue of Maxwell's electromagnetic stress tensor. Note that tii satisfies
the identity
at ii =
-
1 au
-
+ - 8. U . ( 128)
axi -
4nG ax i
Now if we integrate eq . ( 125) over any volume, V, containing the ath body
and no other body in the system, the second and third terms of the left-hand
side will give no contributions because of Gauss's theorem and the vanishing
of p and p outside the body. Using eqs. (7)-(8) we therefore obtain the
following equations of external motion :
( 129a)
where the external force is given as an integral over any surface S enclosing
the ath body (and no other body) :
Fi,1 = -
I. d S/i . ( 129b)
Eqs. ( 128)-(129) are the Newtonian analogues of the famous Einstein

Infeld-Hoffmann surface-integral approach to the problem of motion in
general relativity. Note that because of Gauss's theorem and the identity
( 128) the value of the surface integral ( 129b) is independent of the choice of
the surface, S, as l ong as it stays outside the matter (so that U satisfies
Laplace's equation).
Now the expression ( 129b) for the external force depends only on the
gravitational field outside the bodies, say uout . From eq. (40) uout can be
176 T. Damour
written as :
( 1 30)
with
( 1 3 1)
and
( 1 32)
(actually we are now cheating a little because we are using the vanishing of
the dipole term in uou t which, however has been proven only by an
'
argument using some knowledge of the internal structure). The effacement
of internal structure in the equation for the external motion ( 129) is now
clearly apparent if we choose for the arbitrary surface S a surface of
dimensions - R, the interbody separation, and if we use (for a while) the
external units (eq. (34)). Then both um and um, i are of (numerical) order
unity, while UQ and UQ, i are of order cx 2 . Therefore
FteJ = F� + O(cx2), ( 1 33)
where
( 134)
is of numerical order unity, with
1
5' um um ]
1
tml.J. ·.-
- +
-- [ um , i um ,j zuii
_ _
•k ,k ( 13 5)
4nG •
Therefore the preceding surface-integral approach allows one, without any

calculation , to conclude that, up to order cx 2 , the equations of motion
involve only the centres of mass and the masses (because F� cannot depend
on anything else). However, we need an explicit calculation to work out F� .
There are several ways of doing this; let us spell out one way which has been
found useful in the relativistic generalisation of this approach.
First we notice that, although in the first formulation ( 129b) it was
necessary for S to stay outside the ath body, therefore a finite distance away
from za, in the final (approximate) formulation ( 1 34) S can be any surface
containing the point z0 (and no zbs, b # a) because of Gauss's theorem and
the identity ( 128). In particular, if convenient, we can let S shrink towards za
without changing the value of F�. But then we see that t� will tend to infinity
because of the singularity of um near x = za . Evidently, it is guaranteed a
priori that F� will be finite and that the pole-like singularity is not 'physically
real'. But still, at the computational level , past experience in many fields of
physics has shown that it is efficient to use some ideali sed mathematical
tools (point particles, delta functions , plane waves, . . . ) to deal with real
physical situations. However, one then needs to have a clear mathematical
framework to use such tools (distribution theory, Fourier transform, . . . ). In
our case distribution theory is not appropriate to deal with t� which is
quadratic in um . we can however use the following trick (inspired by the
' '
work of M . Riesz, 1949). Let us consider, if we are studying the motion of the
ath body,
( 1 36)
where ra = = I x - ia(t )I , A is a complex number, and
'°' Gm b
�)a = = ( 1 37)
/ta I x - ib ( t) I '
Now let us define, for a given surface S, the following complex function of the
complex variable A :
F�(A ) '= -i dS/�(A), ( 1 38)

with
; · A == [
1 a u� a u� 1 b a u� a u�
4nG ox ; oxi - 2 ii oxk oxk
t,;!( ) . ( 1 39)
]
It is easily verified that F�(A) is differentiable with respect to A (because ra is
always non-zero on S) and that
F�(O) = F� . ( 140)
(Note that for A =J:. 0, F�(A) depends also on the choice of S.)
lf we consider now t�(A) as a field defined within S (and not only on S), it is
a smooth function of the field point, x, except perhaps at x = ia near which,
usmg
Or A- I
a = (A - l)(x ; - zai ) raA - 3 ' ( 14 1)
ox '
.
it is seen to behave like

(x i ia ) (xi
- - za a
i ) ,_: A - 6 . .
a -4 ·
_ 1-2 b 11 r:A
Hence if the real part of A is large enough (for instance, Real(A) > 3) t�(A)
will define a field which is differentiable at all points within S, including
x = ia . Then for such large A we can apply Gauss theorem to transform the
surface-integral expression ( 138) into an integral over the volume V
178 T. Damour
i
contained within S :
o tii (A)
pi (A) = - 3
d x m ( 142)
v
•
m
.
ox'
Now the identity ( 128) also holds for t�(A) ; hence

o t�(1) 1 ou
= -- � ur;: 1
[ ]
ox' 4nG ox'
m L\ - l G or: - 1 o U�)a
= a ( raA ) ma ( 143)
oxi + ox'
'
4n
L\
where we have used U�)a = 0 within V.
Using also
( 144)
we see that if we develop the right-hand side of eq. ( 143) the first term will
contribute to F�(A) the quantity :
-I d 3x �:,; A(A -
1) 2 (x' z�)r; A • ,
- - ( 145)
which is clearly zero by symmetry arguments if we choose for S a sphere of

radius Ra centred on .ia (the method works as well for arbitrary surfaces but
we wish to simplify the presentation). We are left with
m
pi (A) = - a A(A - l)
m 4n
iv
d 3 xraA - 3
o um)a .
(�
ox'
( 146)
Both the left-hand and the right-hand sides of eq. ( 146) are analytic functions
of A in the domain Real(A) > 0 (such that the volume integral is convergent).
As they coincide for Real(A) > 3 , as has been just shown, they will also
necessarily coincide for Real(A) > O. Now when A � 0, the factor in front of
the right-hand side of eq. (146) shows that only the infinitesimal
neighbourhood of x = .ia contributes to the limit F�(O) = F� (see ( 140)).
Plugging into eq. (146) the Taylor expansion of U� )a ,i around x = .ia it is
easily seen that only the first term contributes to the limi t :
m m A�o
{
Fi (O) = Fi = limit - � A(A
41t
m - l)
0 um)a
;::i 1
ux
(� (.ia )
iR. 4nr2a dra raA - J }
0
( 147)
as shown by an easy calculation.
The final result , combining eqs. ( 128), ( 133), ( 137), ( 147), is

2 .
d z� o Gm 2
ma 2 - ma L � 1 · - b· 1 + O(a ) ,
(
u za Za zb ) ( 148)
_
dt h #- 1.1
in agreement with eq . (4 1).

In summary , the pith of this 'complex analytic continuation' consists of
replacing the initial computation of the 'real' total force acting on the ath
body,
Fai =lv. d3xp U = --4nG1 - 1v. d3x 11. U U .

•1
.
,1 •
( 149)
where p is the 'real ' density and U the real total gravitational potential
(self+ external), by the computation of the analytic continuation to A = O of
F� ( A ) = -
4 G
� L d 3x t.ir;;u;:, , , ( 150)
which contains an A-modified gravitational potential associated with the

'point-mass' ( za , ma ). But , although, computationally speaking, this method
can be seen as a 'point-mass + analytic regularisation' method , we recall that
the real origin of this method lay in the surface-integral ( 129b) of the 'real '
gravitational potential outside the body. A generalisation of this method to
the Einsteinian case has recently been used in the computation of the
equations of motion of two condensed bodies (Damour, 1983a). And
similarly, although the method resembles a 'regularised point-mass'
calculation , it is in fact based on a rigorous surface-integral formulation of
the external equations of motion essentially due to Einstein, Infeld and
Hoffmann (1938) and to Kerr ( 1959).
The lesson of the preceding discussion is that both in Newton's and in
Einstein's theory we can replace the usual Newtonian view of the
gravitational interaction as a set of forces acting on , and being produced by,
each element of matter, by a view, say a la Faraday, where the gravitational
field between the bodies contains a stress distribution which acts on the
bodies via its integrated effects on surfaces enclosing them . The advantage of
the second formulation is that it allows one to prove more simply (because
one does not need to look at the detailed cancellation of self-forces), and in a
more general case (allowing strongly self-gravitating bodies instead of the
only weakly self-gravitating bodies which are within the reach of post
Newtonian methods), the property of effacement of internal structure in the
general relativistic external problem of motion (for different
implementations of this approach 'from outside', see e.g. D' Eath, 1975b;
1 80 T. Damour
Damour, 1983a ; Thome and Hartle, 1985, and references therein to the
earlier works of Weyl, Einstein and others).
6. 15 The problem of gravitational radiation damping and the

relat'i vistic Laplace effect
The discovery of the binary pulsar 1 9 1 3 + 16 (Hulse and Taylor, 1975), a
seemingly clean system of two condensed objects swiftly orbiting around
each other, has raised the hope of observing effects due to gravitational
radiation damping. Indeed , the measurement of the arrival times on Earth of
the radio pulses emitted by one of the objects (the pulsar) allows one to track
the orbital motion of the pulsar with remarkable accuracy (now achieving
+ 6 km both on short time scales, 5 min ; and on very long ones, ten years) .
Therefore, even very small damping forces in the system will become
observable by their secular effects on the orbital motion. Such an effect has
indeed been observed by Taylor and coworkers (Taylor, Fowler and
McCulloch, 1979; Taylor and Weisberg, 1982) with sign and magnitude in
excellent agreement (now better than 4 per cent) with the so-called
'quadrupole formula', a formula which had been previously derived (Peters
and Mathews, 1963 ; Peters, 1964) by inferring the secular effects on the
orbital motion of the loss of energy in the form of gravitational radiation.
This suggests the conclusion that, for the first time, a clear proof of the
existence of gravitational radiation, with the properties predicted by
General Relativity, has been obtained .
Because of the importance of this conclusion one must critically examine
both the hypothesis and the chain of deductions leading to it. We shall not
examine here the hypotheses concerning the structure (two condensed
bodies) and the 'cleanness' (no other effects perturbing the orbital motion) of
the system (see Will , 198 1 ; Taylor and Weisberg, 1982; Damour and
Deruelle, 1986). Instead we shall concentrate on the theoretical arguments
used in the derivation and application of the 'quadrupole formula'.
We do not wish to review in detail the controversy about the meanings
and the derivations of several 'quadrupole formulae' and the related
problem of gravitational radiation damping (see e.g. Ehlers et al. , 1 976 ;
Damour, 1983a; Walker, 1984; Ehlers and Walker, 1984; Schutz, 198 5 ;
Will, 1986, and references therein). However, we wish t o emphasise that
most of the work which has been done concerning the quadrupole formulae
has only an indirect bearing on the problem of explaining the secular
acceleration of the orbital motion of PSR 19 13 + 16 observed by Taylor and
coworkers. Indeed , one class of quadrupole formulae gives information
about a 'gravitational wave energy flux', say F, in the wave-zone. However,

this outgoing flux has never been convincingly related to the time derivative
of some 'local energy of the system' given as an explicit functional of the
instantaneous state of the system with an accuracy comparable with the (very
small) outgoing flux. Indeed , if we know nothing about the functional
dependence of some 'energy of the system', E, the equation ,
=
dE
dt
-
F '
( 1 5 1)
says nothing about the evolution of the state of the system (see e.g . sections
13 and 1 5 of Damour, 1983a , for a more precise discussion of this point).
Moreover, even if we had in hand a well-defined (and well-proven)
conservation law of the type ( 1 5 1), we would have still to face the problem
that all derivations of eq . ( 1 5 1) deal with the total energy flux, F, and the
total 'energy of the system', E. Now in the case of the binary pulsar there is,
roughly speaking , as much energy flux in the form of electromagnetic waves
as there is in the form of gravitational waves, and as much spin kinetic
energy as orbital energy . Therefore, a total (non-detailed) energy balance
equation ( 1 5 1) says, a priori, nothing about the orbital evolution of the
system. A second class of quadrupole formulae gives information about the
presence of 'time-odd' terms in the local equations of motion (terms that
change sign under time reversal). However, the mere knowledge of the
presence of these terms in the equations of external' motion,
:i = A
d e even
(X e ) + Aodd(Xe ), ( 1 52)
does not allow one to conclude that the type of 'secular' effects which are
observed must come only from A odd . Indeed, A even could induce time
symmetric effects of long period which, on the time scale of the observations
(ten years) could look like real secular effects (there are many instances of
this in the mechanics of the solar system).
Therefore, in order to be able to meaningfully compare the theory and the
observations it is necessary to derive the full equations of external motion for
a binary system up to the appearance of time-odd terms, and to control their
solution by a method powerful enough to clearly exhibit all secular (or long
period) effects in the orbital motion . In other words, what is needed is a
complete relativistic celestial mechanical description of a binary system.
Few works have attempted such a complete description . Many works on
radiation damping (e.g. by Anderson and Decanio, Kerlick, Papapetrou
and Linet, Futamase, . . . ) have stayed at the level of a global
1 82 T. Damour
hydrodynamical description without explicitly � onsidering a system of N

separated bodies. Other works (e.g. by Breuer and Rudolph, Linet, . . . ) have
considered a system of N separated bodies but have only partially computed
the equations of orbital motion of the bodies. References to these works will
be found in Damour (1983a), Walker ( 1984), Ehlers and Walker (1984),
Schutz ( 1985), Grishchuk and Kopejkin (1986). Here we shall focus our
attention on three different lines of work which have sought a complete
kinematical description of a two-body (or N-body) system up to the level
where radiation damping first appears.
On the one hand, a new method has been set up to obtain the equations of
external motion of two condensed bodies (Damour, 1980, 1983a). This
method combines the best tools now available to treat the problem : (1) a
post-Minkowskian approximation scheme to obtain the gravitational field
outside the bodies incorporating a natural 'no-incoming-radiation
condition' and giving rise to no 'near-zone only' problems, (2) a matched
asymptotic expansions scheme used to prove the property of effacement
(see Section 6. 14) and to uniquely determine the gravitational field external
to N condensed bodies, and (3) an Einstein-Infeld-Hoffmann-Kerr-type
approach to compute the equations of orbital motion from the knowledge of
the external field only (the explicit calculation of the surface integrals being
simplified by the use of the analytic continuation trick explained in Section
6. 14). Then this method has been implemented, in the case of two bodies, to
compute the external gravitational field (Bel et al. , 198 1 ; Damour, 1983c)
and the equations of orbital motion , obtained first in retarded integro
differential form and then expanded in vext;c to get a conventional
differential system. The final equations of motion (at the 2!-th post
Newtonian level) are expressed only in terms of the instantaneous positions,
velocities and spins in a given harmonic coordinate system. 'Time', t, means
here the zeroth harmonic coordinate of the post-Minkowskian external
scheme . 'Position', z, z' , means the spatial harmonic coordinate of each
'centre of field', 'velocity' means vi : = dzi/dt, v' i := dzt i/dt. 'Spin', Sii , S�i'
means the spatial components of the antisymmetric (Minkowskian) tensors,
Sµv • S� v describing in the post-Minkowskian-external-field approach the
intrinsic angular momentum of the bodies. They are constrained to satisfy
the following Minkowski-covariant spin condition :
( 1 53)
Let us introduce the notation., RNi := zi (t) - z'i(t) (w.i th NiN i = 1), Vi : =
vi (t) - v' i (t), (N v) := N i vi (etc . . . ), v 2 := vivi (etc . . . . ) and a i := d vi/d t for the
.
accelerations. Then each body must satisfy the following equation of motion
(Damour and Deruelle, 198 la ; Damour, 1982) :
ai = A�(i - i') + c - 2 A�(i - i' , V, v') + c - 4 A�(i - i' ' v, v' ' S, S')
+ c- 5 A�(i - z', v - v') + O(c - 6), ( 1 54)
with
A� = - Gm'R - 2 Ni, ( 1 55)
A � = Gm' R - 2 {Ni[ - v 2 - 2v' 2 + 4(vv') + �(Nv') 2 + 5(Gm/R) + 4(Gm'/R)]
+ (vi - v' i ) [4(Nv) - 3(Nv')] } , ( 156)
A� = B� + C� + D � , ( 1 57)
B� = Gm' R - 2 {N i [ - 2v' 4 + 4v' 2 (vv') - 2(vv') 2 + �v 2(Nv') 2 + -!v' 2 (Nv') 2
- 6(vv')(Nv') 2 - 185(Nv') 4 + (Gm/R)( - 1fv 2 + �v' 2 - 1(vv') + 1
- 39(Nv)(Nv') + \7 (Nv') 2) + (Gm'/R)(4v' 2 - 8 (vv') + 2(Nv) 2
-4(Nv)(Nv') - 6(Nv') 2)]
+ (vi - vi i) [v 2 (Nv') + 4v' 2 (Nv) - 5v' 2 (Nv') - 4( vv')(Nv)
+ 4(vv')(Nv') - 6(Nv)(Nv') 2 +-!(Nv') 3
+ (Gm/R)( - 6l(Nv) + 545(Nv')) + (Gm'/R)( - 2(Nv) - 2(Nv'))] } , ( 158)
C� = G3m' R - 4 Ni [ - 547 m 2 - 9m' 2 - 6/mm'J ,
) ( )
( 1 59)
(
m
ik S' ik
m
)
D � = S + 2 , (vi - v' 1 ) ( ) (
G m'
R ,kt
+ 2
5k1
m
+ 2
S'�1 1 1 Gm'
m
(v - v' )
R , ik '
(160)
and
A� = 1G 2 mm'R - 3{ Vi [ - V 2 + 2(Gm/R) - 8(Gm'/R)]
+ Ni(N V)[3 V2 - 6(Gm/R) + 5l(Gm'/R)] } . ( 16 1)
The two parameters m and m' appearing in eqs. ( 154)-( 16 1) are the
'Schwarzschild masses' of the condensed bodies. They are two constants
which appear in the external gravitational field, in which are hidden many
internal structure effects (see the discussion of the 'effacement of internal
structure' in Section 6. 14). On the other hand, the spin tensors undergo a slow
evolution (on the post-Newtonian time scale, i.e. Pe- 2 times the orbital
period) which is also obtained in the Einstein-Infeld-Hoffmann-Kerr-type
approach (Damour, 1982, and references therein). Introducing, a la Schiff, a
suitable spin-vector, S , associated with Sµv • the law of evolution ('spin
precession ') reads for the first body (see also references in Section 6 . 1 3 .2)
dS [ Gm' - 3
- = -- N x - v - 2v
dt c 2R 2 2
( +')] x -S + O (-c14)
+
. ( 162)
1 84 T. Damour
The following equations of orbital motion for a binary system , obtained

from eq. ( 1 54) by leaving out the spin-orbit coupling term and the term of
order c - 5:
{a i = di (z, z' , v, v') := A� + c - 2 A � + c - 4 (B� + C�), ( 163a)
( 163b)
a' i = Jil ' i (z, z' , v, v') : = A� + c - 2 A 1 + c- 4 (B� + C 1 ),
have been explicitly shown not to be deducible from a conventional
Lagrangian (function of positions and velocities) but, instead, from a
generalised Lagrangian (depending on accelerations) (Damour and
Deruelle, 198 lb; Damour, 1982) :
L(z, v, a) = L0 (z - z', v, v') + c - 2L 2 (z - z' , v, v') + c - 4L 4 (z - z ' , v, v' , a, a'),
( 1 64)
( 165)
and
1 (
L 2 = L 8 mv4 +
Gmm' 3 2 7
R 2 v - 4
[
(vv ) -
1
4 (Nv)(Nv
,) -
1 Gm
2R
, ]) . ( 1 66)
L4 (z, v, a) = M 4 + N 4 , ( 167)
M 4 (z, v, a) = L(/6 mv 6 ) + LGmm' R - 1 [iv4 + ��v 2 v' 2 - 2v 2 (vv') + !(vv') 2
- i(N v) 2 v' 2 + i(N v)(N v')(vv') + 136 (N v) 2 (N v') 2]
+ LG 2m 2 m' R - 2 [-!v 2 + iv' 2 -i(vv') + 1(Nv) 2
+ �(Nv') 2 - -!(Nv)(Nv')]
+ LGmm' [(Na)(iv' 2 -! (Nv') 2 ) -i (v' a)(Nv')] , ( 168)
and
( 169)
The generalised Lagrangian ( 164) is (approximately) invariant under the

Poincare group. This allows one to construct ten Noetherian quantities
which would be conserved during the motion ( 163) (Damour and Deruelle,
198 1c; Damour, 1983a), EN oei , pNoei, JN oet , J(Noei , which are single-valued
functions of the instantaneous state of the system (they can be reduced so as
to depend only on simultaneous positions and velocities).
Now if we go back to the more exact equations of motion ( 1 54) for the
{
binary system, they read :
d = d i + c - 4D� + c - 5A� , ( 170a)
a' i = d l i + c - 4D1 + c - 5A'� , ( 1 70b)
They can be solved by treating c - 4D� + c - 5 A� as a perturbative force
superimposed on the Lagrangian system ( 163). The use of the method of

variation of arbitrary constants allows one to control the effects of these
perturbative fo rces in great detail. For the effects of c - 4D� see Ba.rker and
O'Connell ( 1975, 1979), and for those of c - 5 A� see Damour ( 1 983b, 1985).
In particular it has been shown that the 'relativistic mean anomaly' l (an
angle such that IN = 2nN for the Nth return of each body to its (proper)
periastron) has the following (approximate) time dependence :
l _ c0 t 1 P0 2
- +p
. .
p� t + penod1c terms,
0 -2
( 1 7 1)
2n 2n
where the periodic terms have only a short period � P 0 ('orbital period').
This proves that the observable secular effect is given only by the third term
on the right-hand side of eq . ( 17 1 ), where, as shown by a direct variation-of
( )
constant calculation :
. 192n 2nG 513 mm' 1 + Ue 2 + Jie4
Po = ( 1 72)
- Sc"S P o (m + m') l / 3 ( 1 - e 2 ) 7 / 2
e denoting a 'relativistic eccentricity' (for details of the meaning and proof of
eqs. ( 1 7 1}--( 1 72) see Damour, 1985, for earlier heuristic proofs see Esposito
and Harrison ( 1 975) and Wagoner ( 1 975)).
Finally , these results concerning the coordinate motion of the binary system
in a special harmonic coordinate system geared to the binary system have to
be related to the quantities actually measured by the observers. This
problem has been re-examined recently with a view to correcting, simplify
ing and completing previous works (see Damour and Deruelle, 1986, and
.
references therein). This allows one to conclude that the 'theoretical
parameters' introduced above (such as m, m' , P 0 , e, P 0 , . . ) coincide with
their corresponding 'observed parameters', denoted mP , me, Pb , e, Pb , . . . .
On the other hand, a recently developed program (Schafer, 1983 , 1985,
1986) based on a Hamiltonian approach to the interaction of N spinless
point-particles with the gravitational wave-field has allowed one to reach
elegantly (in a different gauge mo re adapted to the separation of
conservative and damping effects) the main result of the previous method :
i.e. a force, equivalent to c - 5 A� (coming from the interaction with the
dynamical degrees of freedom of the gravitational field) acting on the
Hamiltonian sub-system of the instantaneously interacting N particles.
However, the treatment of the problems associated with the use of point
particles has not been fully justified so that the physical relevance of this
work is not obvious on its own.
Another line of work aimed at treating the problem of the motion of a
1 86 T. Damour
binary system with the same completeness as the ·first method discussed
above, has also been recently reported (Grishchuk and Kopejkin, 1983,
1986; Kopejkin, 1985). It is based on : ( 1) a post-Newtonian approximation
scheme of the type of eq . (94), and (2) the assumption that the bodies are
non-rotating, 'spherically symmetric' fluid balls. Spherical symmetry is
taken in the coordinate sense, so that the centre of mass defined by eq . ( 1 1 5),
with p given by eq . ( 1 17), coincides with the centre of symmetry. Then the
equations for the motion of the centre of mass of each body are obtained by
integrating the local post-Newtonian equations of motion (69b). They have
been explicitly calculated, retaining all the higher derivatives that appear. If
one 'reduces' these higher derivatives by using the lower-order equations of
motion (as was done in the first method) the explicit result of this post
Newtonian calculation is (Kopejkin, 1985) :
(µ + c - 2µ 2 + c - 4µ 4)a i = [(µ + c - 2µ 2 + c - 4µ 4)A�(µ' + c - 2µ� + c - 4µ�)]
+ c - 2 [(µ + c - 2 µ 2 )A�(µ + c - 2µ 2 , µ ' + c - 2 µ� )]
+ c - 4µ(B� (µ, µ') + C�(µ, µ'))
+ c - 5µA � (µ, µ') + O(c - 6 ), ( 173)
where the square brackets mean that one should expand in exp lic i t powers
of c - i and discard all powers greater than or equal to the sixth. In eq . ( 173)
the parameter µ denotes the 'rest mass' of the body :
µ •= L d 3xr. , ( 1 74)
while
µ 1 •= fv d 3 xr . (TI 0 - 1U1'lj, ( 1 75)
µ 4 '= L d 3 xr . ul•l(!Ul•l _ TI 0 - 3p 0/r .) . ( 1 76)

Now , formally speaking , if we set
m ==µ + c - 2 µ 2 + c - 4µ 4 , ( 1 77)
and if we forget about the truncation implicit in the square brackets, eq .
( 1 73) becomes, identical , modulo O(c - 6 ), with eq . ( 1 54) minus the spin-orbit
term ( 160). This possibility of sweeping the various relativistic corrections,
c - 2µ 2 and c - 4µ 4 , into the 'effective' mass ( 1 77) is a direct post-Newtonian
proof of the effacement of internal structure in the external motion, and was
discussed in the preceding section. Finally, the kinematics of the binary
system has been studied along the same lines as the variation-of-constant
approach sketched above. In particular, Grishchuk and Kopejkin ( 1986)

also introduce a generalised Lagrangian function of positions, velocities and
accelerations. Their Lagrangian is a member of the general class of
generalised Lagrangians described in Damour and Deruelle ( 198 1b), i .e.,
d
15en (z, v , a) = L(z, v, a) + {F (z , v)} + c - 4Q(a - A 0 ), ( 178)
dt
where L(z, v, a) is the Lagrangian ( 164), d F/d t = oF/oz v + oF/ov · a is a total
·
time derivative which does not contribute to the equations of motion, and
Q (a - A 0 ) is a quadratic form of the differences a - A0 (z, v), a' - A0(z, v) which
does not contribute to the order-reduced equations of motion at the order
considered . They would contribute at the order c - 6, but then it is sufficient
to replace a - A 0 by a - A 0 - c - 2 A 2 to postpone the influence of such
'double-zero' terms (using the terminology of Barker and O'Connell ( 1980)).
The statement of Grishchuk and Kopejkin that the Lagrangian (164) is 'only
valid in a non-harmonic frame of reference' is incorrect and is based on a
confusion between the effects of a 'double-zero' term such as c - 4Q(a - A 0 )
(which does not contribute) and those of what one might call a 'simple-zero'
term , i.e. a linear function, c - 4.P(a - A 0 ), of the differences a - A 0 . As
remarked recently by Schafer ( 1984), a simple-zero term does contribute to
the equations of motion but in a way which is equivalent to a change of
coordinate system at order c - 4. For a general investigation of the effect of
infinitesimal coordinate transformations on generalised Lagrangians see
Damour and Schafer ( 1985) and references therein. In particular, it has been
shown in the latter work that the acceleration-dependent two-body
Lagrangian ( 164) is equivalent, modulo a suitable coordinate
transformation, to the special N = 2 case of an ordinary N-body Lagrangian,
L(z, v), derived earlier by Ohta, Okamura, Kimura and Hiida ( 1974b), who
were using a non-harmonic coordinate system.
The fact that two independent methods (post-Minkowskian + Einstein
Infeld-Hoffmann and post-Newtonian + perfect fluid) give formally
identical equations of motion, in harmonic coordinates, is a strong
confirmation of the validity of the numerical coefficients in eqs. ( 155)-( 159)
and ( 16 1). So much so that the coefficients in ( 165)-( 169) are further
confirmed by comparison with the works of Ohta et al. ( 1974a, b) and
Schafer ( 1985) based on a third independent method (post-Newtonian +
delta functions) (after correction of a wrongly evaluated two-body integral
in Ohta et al. ( 1974a) ; see Damour ( 1983a, section 13); Damour and Schafer
( 1985)). However, only the first method has been tailored to deal with the
188 T. Damour
astrophysical problem at hand, i.e. the non-linear-retarded gravitational

interaction between two condensed objects. Indeed, the first method, as
sketched above, essentially uses only one small parameter · which can be
taken as 'Ye = Gm/c2 R. The parameter rx = L/R is of the order of Ye because L
Gm/c2, by definition of condensed bodies, and the parameter fle is used only
in the last step of the method and is then related to '}' e by fle ,.., '}': 12 . By
construction tl;iis method does not require that the internal gravitational
field be weak ; on the contrary, it uses l' i "' 1 to simplify things (however, it has
also been assumed that /Ji2 ;S 'Ye in order to be able to efficiently use the
effacement property).
On the other hand, the third method sketched above essentially uses the
usual post-Newtonian + perfect fluid approach. This means : ( 1 ) that there
are problems with incorporating boundary conditions (which show up as
divergent integrals in the next approximation), and (2) that it can be applied
only to weakly self-gravitating bodies. We have even shown in Section 6 . 1 3 .2
that, at the level at which explicit calculations have been performed ,
retention of the 2 and 2. 5PN terms in the external equations of motion was
logically inconsistent because the uncontrolled error term formally denoted
'O(c - 6 ) ' in eq . ( 173) was in fact at least - y: si s , i .e. larger even than the 2PN
corrections. This means that application of the PN equation ( 1 73) to the
binary pulsar cannot, at present, be justified . The fact that the 'sweeping' of
the self-field effects into the effective mass ( 177) works at the 2PN level is an
indication, but certainly not a proof, that eq . ( 173) can be applied to strongly
self-gravitating objects. In fact, the argument above says that , at present,
there is no proof that eq. ( 1 73) can be meaningfully applied to any objects at
all ! Note also that the divergence problems appearing at the 3PN
approximation (see e.g. Kerlick, 1980), which is an indication that the 3PN
error is bigger than expected , can only strengthen our argument.
Finally, it should be emphasised that this issue of strong self-field effects is
not only important from the theoretical point of view of the justification of
the derivation of the equations of motion , but it is also essential from the
purely numerical point of view. Indeed, in the case of the binary pulsar one
expects self-fields 'Yi - 10 per cent so that, for instance, the last term of eq .
( 173) as obtained explicitly by the post-Newtonian method , c - 5 µA � (µ, µ'),
differs by - 10 per cent from the corresponding explicit result of the previous
method , eq . ( 16 1), because of the difference in mass (m - µ)/µ - Yi · Now the
variation-of-constants-calculation sketched above shows that the observed
effect ( 172) comes directly from the term of order c - 5 in the equations of
motion . Therefore, if one takes at face value the equations of motion ( 1 73) as
they come out of the post-Newtonian calculation they predict a . secular

acceleration of the mean anomaly which differs by , say, more than 10 per
cent from the prediction ( 1 72 ) of the other method (consistently taking into
account that within all methods the observed masses should be identified
with the ones appearing in the lPN equations of motion, i .e. m in the first
method, and µ + c - 2µ 2 in the third method). Now the observed value of P
coincides with the predicted value to within less than 4 per cent, so that such
a difference in prediction makes the post-Newtonian + perfect fluid
approach, truncated at the 2PN level , of dubious use from the numerical
point of view .
In concluding this section , we can say that, in General Relativity, the
2.5PN equations governing the orbital motion of two condensed bodies can
be considered as a perturbed Lagrangian system. The Lagrangian system
( 16 3 ) (deducible from the generalised Lagrangian ( 164)) gives rise, in the
bound case, to a multi-periodic motion (with a radial period � P 0, and an
angular period ,...., P of3e- 2 , 'periastron precession'). Then the most interesting
perturbation is the last term of eq. ( 1 54) :
·- s ;
·-c - AS • ( 1 79 )
Eimn stei n
da ping
which can rightly be considered as a 'radiation dam P,ing' term because it
leads to a secular decrease of the Noetherian energy of the 'unperturbed'
Lagrangian (and because, in Schafer's approach, it does appear as a direct
coupling with the 'radiative degrees of freedom' of the gravitational field).
Indeed, using
( 1 80)
and ,
- 1.S ziQ ij( S ) + 2 d/(4> S ISS + O(c -
2) '
A ;S - U /(3> + .!J�
ij + 3 ai S ISS> - }�4> (181)
with
and ( 1 82 )
" Gm z 2 zi ,
1;55 := � J.ISS == L Gmz 2 v i ,
one finds
d E N o et
- c S - E s - -- Q 11(. 3. ) Q (11. 3. ) + 0 (c - 6 ) '
- d 1
-- = ( 183)
dt dt 5Gcs
where both EN o et and E s are univalued functions of the instantaneous state
of the binary system : z(t), z'(t), v(t), v'(t). Therefore ENoet experiences a
1 90 T Damour
secular decrease,
(d£Noet ) = l_ ( Q PlQ �?l)
5Gc5 1 1 11 '
_ _
( 184)
dt
whose value coincides with the so-called 'quadrupole flux formula' giving
the outgoing flux of gravitational radiation in the wave zone. Actually, this
'quadrupole flux formula' is, in my opinion , rather less well established than
the 'secular acceleration (quadrupole) formula' ( 1 72) (because of the
problem ofjustifying the 'good' quadrupole ( 1 82) for condensed bodies), but
I quote it here only to justify the name 'gravitational radiation damping'.
However, I wish to stress that what is actually observed is an orbital effect ,
so that Adamping should rather be thought of as the relativistic descendant of
the Laplace effect.
Indeed , Laplace ( 1805), in his famous Traite de M ecanique Celeste which
studied 'les alterations que le mouvement des planetes et des cometes peut
eprou ver [ . . . ] par la transmission successive de la pesanteur' (i .e. the orbital
perturbations due to a finite velocity of propagation of gravity) , introduced
the idea that if gravity propagates with a finite velocity, say cg , then there will
be a small modification of Newton's law (due to aberration effects)
amounting to the addition of a small 'damping' term of the type :
AidLaplm aiceng - - Gm' Vi

a p R2 � ·
( 1 8 5)
He then showed that this damping term would cause a shrinkage and a
circularisation of the orbit, together with a secular acceleration of the
angular motion . He was aware of the fact that the last effect is the easiest to
observe; and even concluded from the 'known' secular acceleration of the
Moon that the velocity of propagation of gravity must be
c � 7 x 106
g c1ight ! ( 1 86)
The last conclusion of Laplace is not quite correct, as we now know , but
we can still admire his profound insight into what would be the main orbital
effects caused by a finite velocity of propagation of gravity. Indeed, it is clear
from eq. ( 16 1 ) that the Einsteinian damping term ( 179) (in harmonic
coordinates) contains many terms of the Laplace damping type ( 1 8 5).
However, the particular structure of Einstein's theory has resulted in
considerable reduction of the relative magnitude of this damping term from
the (not so) 'naive' expectation , /c , due to aberration effects , down to the
V
present ( /c ) 5 Still, the qualitative conclusions of Laplace are correct, the
V .
main effects of the damping term ( 1 79) being, indeed , a shrinkage and a
circularisation of the orbit (Peters , 1 964) and a secular acceleration of the

mean anomaly given by the eqs. ( 17 1 )-( 1 72) (which can be called the 'secular
acceleration quadrupole formula' because its demonstration shows that P 0
is proportional to the secular decrease of £Noet which, according to eq . ( 184)
can be expressed in terms of the 'quadrupole moment' , Q ii• of eq . ( 182)).
Such an acceleration effect has indeed been observed in the binary pulsar
PSR 19 13 + 16 with a sign and a magnitude which is in remarkable
agreement with the theoretical formula ( 172) (within 4 per cent, Weisberg
and Taylor, 1 984). This excellent agreement provides an impressive
confirmation of the structure ( 170) of the equations of motion. As this
structure was derived from eqs. ( 1 54)-( 16 1), which were themselves derived
by using both the fully non-linear structure (strong self-gravity, effacement
property) and the radiative structure (retarded integrals) of Einstein's
theory , we can say that this constitutes a profound confirmation of the non
linear hyperbolic structure of Einstein's theory and thereby an indirect proof
of the existence of gravitational radiation. M oreover, some (semi-heuristic)
investigations within other relativistic theories of gravity (see Will, 198 1 ,
and references therein) indicate that this is also a very sensitive test fo r a
gravity theory which might allow us to discard many alternative theories
which, like Einstein's, are compatible with the weak-field quasi-static
relativistic effects in the solar system.
6. 16 Conclusion
Henri Poincare ( 1934) once remarked that real problems can never be
classified as solved or unsolved ones, but that they are always more or less
solved ('ii y a seulement des problemes plus ou moins resolus'). This remark
applies particularly well to the problem of motion which has had a
chequered history . Even the Newtonian problem of motion, which appeared
to be well understood after the development of the powerful methods of
classical celestial mechanics (codified around 1889 in the treatise of
Tisserand), embarked on an entirely new career after the work of Poincare
( 1892) which has led to many further developments (see e.g . Arnold, 1978;
Gallavotti , 1983). In my opinion the Einsteinian problem of motion has not
even reached a classical stage where the basic problems appear
(provisionally) as 'well understood'. At first sight the best-developed
approximation method in general relativity, the 'post-Newtonian' one,
would seem to constitute such a classical stage, but the literature on the
post-Newtonian problem of motion is full of repetitions, errors or
ambiguities. Several problems (especially at the lPN approximation) have
192 T. Damour
been done over and over again with slightly modified approaches, and
different notation (or worse, the same notation for different quantities), and
still it seems to me that some of the basic issues have not really been
tackled. Moreover, the development of other approximation methods
('post-Minkowskian', 'singular perturbation') has rather complicated the
picture by giving rise to many hybrids (some of which have been discussed
above).
My original intention was to conclude this survey by giving a list of the
issues that need to be clarified. I renounced this project because, if one
wishes to look at the work done with a critical eye, nearly all aspects of
the problem of motion need to be thoroughly re-investigated for
mathematical, physical or conceptual reasons ; so that the list of open
problems would, consistent with the remark of Poincare, include all the
issues discussed above. Another reason for renouncing this project is that
Brumberg (1986) has very recently given such a list, to which I gladly refer
the reader.
One thing is certain : the Einsteinian problem of motion is no longer a
purely theoretical problem , thanks to the dramatic improvement in the
precision of position measurements in the solar system, and to the discovery
of the binary pulsar 19 13 + 16 which is a marvellous relativistic laboratory ;
the Einsteinian problem of motion has become an important tool of modern
astrophysics. It is therefore of some urgency, not only to complete and unify
the work already done, but also to develop new approaches which will keep
the best aspects of existing methods while freeing themselves from their
conceptual and/or technical drawbacks. These new approaches should aim
at both formal and conceptual clarification of the basic issues, and at
securing more accurate explicit results. Let us hope that such methods will
be developed before the (first) centenary of the publication of Einstein's
classic paper on 'the foundation of the general theory of relativity' ('Die
Grundlage der allgemeinen Relativitatstheorie', Einstein, 19 16a) which is
comparable with Newton's Philosophiae Natura/is Principia Mathematica
both in its importance for physics and its structure which consists, after a
conceptual introduction, of a formal mathematical part followed by a
physical part where connection is made with the natural world.
References
Anderson, J . L. ( 1 980) . Phys. Rev. Lett., 45, 1745-8 .
Anderson , J . L. and Decanio, T . C. ( 1975). Gen. Rel. Grav., 6, 197.
Anderson, J . L . , Kates, R. E., Kegeles, L . S . and Madonna, R. G. (1982). Phys. Rev.,

D25, 2038-48 .
Anderson, J. L . and Madonna, R. G. ( 1983). Gen. Rel. Grav., 15, 1 12 1-9.
Arnold , V. I . (1978). Mathematical Methods of Classical Mechanics. Springer : New
York .
Amowitt, R . , Deser, S . and Misner, C. W. ( 1962). In Gravitation: An Introduction to
Current Research, ed . L . Witten , pp. 227-65. Wiley : New York .
Ashby, N . and Bertotti, B . ( 1984). Phys. Rev. Lett. , 52, 485-8.
Ashby, N. and Bertotti , B . (1986). Phys. Rev. D., 34, 2246-59 .
Bailey , I . and Israel, W . ( 1980). Ann. Phys. (N. Y.), 130, 188-2 14.
Barker, B . M. and O'Connell, R . F. (1975). Phys. Rev., D 1 2 , 329-35.
Barker, B . M. and O'Connell, R. F. ( 1979). Gen. Rel. Grav., 1 1 , 149-75.
Barker, B . M. and O'Connell, R . F. (1980). Can. J. Phys., 58, 1659-66 .
Bel, L . , Damour, T . , Deruelle, N . , Ibanez, J . and Martin, J . (198 1). Gen. Rel. Grav., 13,
963- 1004 .
Bel, L . and Martin, J . ( 1981). Ann. Inst. H. Poincare (Phys. Theor.), 34 A , 23 1 and
references therein.
Bennewitz, F. and Westpfahl, K. ( 1971). Commun. Math. Phys., 23, 296-3 18.
Bertotti, B. ( 1954). Nuov. Cim., 12, 226-32.
Bertotti , B . ( 1956). Nuov. Cim., 4 , 898-906.
Bertotti , B . , de Felice, F. and Pascolini, A . , eds. (1984). General Relativity and
Gravitation (GR 10, Padua 1983 ). Reidel : Dordrecht.
Bertotti, B . and Plebanski, J . (1960). Ann. Phys. (N.Y.), 1 1 , 169-200.
Blanchet, L . ( 1987). Proc. Roy. Soc. (Land.), A409, 383-99.
Blanchet, L. and Damour, T. (1984a). C.R. Acad. Sci. Paris (II), 298, 43 1-4.
Blanchet , L . and Damour, T. (1984b). Phys. Lett. , 104A , 82-5.
Blanchet , L . and Damour, T. ( 1986a). Phil. Trans. R. Soc. London, A320, 379-430.
Blanchet, L. and Damour, T. ( 1986b). To be submitted for publication .
Breuer, R . A . and Rudolph, E. (1982). Gen. Rel. Grav., 14, 1 8 1-2 1 1 (this work is an
extension of the work of E. Rudolph and G. Borner (1978a, b), Gen. Rel. Grav. , 9,
809-20 and 82 1-33).
Brillouin, M . (1922). C.R . Acad. Sci. Paris, 175, 1009- 12.
Brouwer, D. and Clemence, G . M. (196 1 ). Methods of Celestial Mechanics. Academic
P ress : New York.
Bru hat, Y. (1962). In Gravitation: An Introduction to Current Research, ed. L. Witten,
pp. 130-68. Wiley : New York.
Bromberg , V . A . ( 1972). Relativistic Celestial Mechanics. Nauka : Moscow (in Russian).
Brum berg, V. A. (1986). In Relativity in Celestial Mechanics and Astrometry (IAU
Symposium 1 14), ed . J. Kovalevsky and V. A. Brumberg, pp. 5- 17. Reidel : Dordrecht.
Burke, W. L. (1971). J. Math . Phys., 12, 40 1- 18.
Caporali, A. ( 198 1a). Nuov. Cim., 6 1 B , 18 1-204.
Caporali, A. ( 198 1b). Nuov. Cim., 61B, 205- 12.
Carmeli, M. ( 1964). Phys. Lett., 9, 132-4.
Carmeli, M . ( 1965). Nuov. Cim., 37, 842-75.
Chandrasekhar, S. (1965). Astrophys. J., 142, 1488- 1 5 12.
Chandrasekhar, S. and Nutku, Y. (1969). Astrophys. J., 158, 55-79.
Chazy , J. ( 1928, 1930). La Theorie de la Relativite et la M ecanique Celeste, vols. 1 and 2.
Gauthier-Villa rs : Paris.
194 T. Damour
Choquet-Bruhat, Y. and Christodoulou, D. ( 1983). In Advances in Mathematics,

Supplementary Volume 8, ed. V. Guillemin, pp. 73-9 1 . Academic Press : New York .
Choquet-Bruhat, Y. and York, J. W. J r ( 1980) . In General Relativity an� Gravitation, ed .
A . Held, vol . 1 , pp. 99- 172. Plenum Press : New York.
Christodoulou, D. and Schmidt, B. ( 1979). Commun . Math. Phys., 68, 275-89.
Contopoulos, G . and Spyrou, N . ( 1 976). Astrophys. J . , 205, 592-8.
Cooperstock, F. I. and Hobill , D . W. ( 1982). Gen. Rel. Grav., 14 , 36 1-78.
Crowley, R . C. and Thorne, K. S . ( 1977). Astrophys. J., 215, 624-- 3 5.
Dallas, S . S . ( 1977) . Ce/. Mech., 15, 1 1 1-23.
Damour, T. ( 1978). In Physics and Astrophysics of Neutron Stars and Black Holes, ed .
R . Giacconi and R. Ruffini (65th E. Fermi School, Varenna, 1975), pp. 547-9. North
Holland : Amsterdam.
Damour, T. ( 1980). C.R. Acad. Sci. Paris, 291 , serie A, 227-9.
Damour, T. ( 1982). C.R. Acad. Sci. Paris, 294, serie II, 1 3 5 5-7.
Damour, T. ( 1983a). In Gravitational Radiation, ed . N. Deruelle and T . Piran (Les
Houches 1982), pp. 59-144. North-Holland : Amsterdam.
Damour, T. ( 1983b). Phys. Rev. Lett., 51, 10 19-2 1 .
Damour, T. ( 1983c) . In Proceedings of the Third Marcel Grossmann Meeting o n General
Relativity, ed. Hu Ning (Shangai, 1982), part A, pp. 583-97. Science P ress : Beijing;
North-Holland : Amsterdam.
Damour, T. ( 1985): In Proceedings of Journees Relativistes 1 983 , ed . S . Benenti,
M. Ferraris and M. Francaviglia (Torino, 1983), pp. 89-1 10. Pitagora Editrice :
Bologna.
Damour, T. ( 1987). In Geometrie et Physique, ed. Y. Choquet, B. Coll , R. Kerner and
A. Lichnerowicz (Joumees Relativistes 1985, Luminy) , pp. 72-83. Hermann : Paris.
Damour, T. and Deruelle, N. ( 198 1a). Phys. Lett., 87A, 8 1-4 .
Damour, T. and Deruelle, N. ( 198 1b). C.R. Acad. Sci. Paris, 293 (II), 537-40.
Damour, T. and Deruelle, N. ( 198 1c). C.R. Acad. Sci. Paris, 293 (II), 877-80.
Damour, T. and Deruelle, N. ( 1986). Ann. Inst. H. Poincare (Physique Theorique), 44 ,
263-92 (this work is a continuation of T. Damour and N . Deruelle ( 1985), Ann . Inst.
H. Poincare (Phys. Theor.), 43, 107-32) .
Damour, T. and Schafer, G. ( 1985). Gen. Rel. Grav., 17, 879-905.
D'Eath, P. D. ( 1975a). Phys. Rev., D l l , 1 387- 1403.
D'Eath, P. D . ( 1975b) . Phys. Rev., D l l , 2 183-99 .
Demianski , M . and Grishchuk, L . P. ( 1974). Gen. Rel. Grav., 5 , 673.
Deruelle, N. ( 1982). These de doctoral d'Etat, unpublished, and ( 1983) in Proceedings of
the Third Marcel Grossmann Meeting, ed . Hu Ning , pp. 955-8 . Science Press : Beijing;
North-Holland : Amsterdam.
Deruelle, N. and Pi ran, T., eds . ( 1983). Gravitational Radiation (NATO ASI , les
Houches, 1982). North-Holland :Amsterdam.
De Sitter, W. ( 19 16). Mon. Not. R . Astr. Soc., 76, 699-728 ; 77, 1 55-84.
Dixon, W. G. ( 1979). In Isolated Gravitating Systems in General Relativity, ed . J. Ehlers
(Varenna, 1976, course 67), pp. 1 56--2 19. North-Holland : Amsterdam .
Droste, J. ( 19 16). Versl. K. Akad. Wet. Amsterdam, 19, 447-55.
Eddington, A. and Clark , G. L. ( 1938). Proc. Roy. Soc . (Lond.), A 166, 465-75.
Ehlers, J . ( 1977). In Proceedings of the International School of General Relativistic
Effects in Physics and Astrophysics, ed . J. Ehlers (Erice, 1977, course 3), pp. 45-73 .
MPI : Miinchen (annotated in 1983).
Ehlers, J., ed . ( 1979). Isolated Gravitating Systems in General Relativity (67th Enrico
Fermi School, Varenna, 1976) . North-Holland : Amsterdam.
Ehlers, J . ( 1980). Ann . N . Y. Acad. Sci . , 336, 279-94.

Ehlers, J . ( 19 8 1 ) . In Grundlagen-probleme der modernen Physik, ed . J. Nitsch et al., pp.
65-84. Bibliogr. Inst . : Mannheim.
Ehlers, J. ( 1984). In Proceedings of the 7th International Congress of Logic, Methodology
and Philosophy of Science. North-Holland : Amsterdam.
Ehlers, J . , Rosenblum, A., Golberg, J. N. and Havas, P. ( 1976). Astrophys. J., 208,
L 77-8 1 .
Ehlers, J . and Rudolph, E . ( 1977). Gen. Rel. Grav., 8 , 197.
Ehlers, J . and Walker, M. ( 1984) . In General Relativity and Gravitation, ed . B . Bertotti et
al. (GR 10, Padua , 1983), pp. 125-37. Reidel : Dordrecht .
Einstein , A . ( 19 16a) . Ann . der Physik, 49, 769 (English translation i n H . A. Lorentz et
al., The Principle of Relativity, pp. 109-64. Dover : New York, 1952).
Einstein, A . ( 19 16b) . Preuss. Akad. Wiss. Berlin, Sitzber., 688-96 .
Einstein , A . and Infeld, L . ( 1940) . Ann. Math., 41 , 455.
Einstein, A . and Infeld, L. ( 1949). Can. J. Math . , 1, 209 .
Einstein , A . , Infeld , L . and Hoffmann, B . ( 1938). Ann. Math., 39, 65- 100.
Eisenstaedt, J. ( 1986) . Arch . for Hist. of Exact Sciences, 35, 1 15-8 5.
Ellis, G. F. R . ( 1984) . In General Relativity and Gravitation, ed . B . Bertotti et al. (GR 10,
Padua, 1983), pp. 2 15-8 8 . Reidel : Dordrecht.
Esposito , L. W. and Harrison , E. R. ( 1975). Astrophys. J., 196, L l-2.
Fock, V . ( 1 959). Th e Theory of Space Time and Gravitation. Pergamon Press : Oxford .
Friedrich , H . ( 198 1). Proc. R. Soc. (Lond.), A378, 40 1-2 1 .
Friedrich, H . ( 1985). Preprint, Hamburg.
Futamase, T. ( 1983). Phys. Rev., D28, 2373-8 1 .
Futamase, T . ( 1 986). 'The strong field point particle limit and the quadrupole formula in
the binary pulsar system'. Preprint , Washington University.
Futamase, T. and Schu tz, B. F. ( 1983). Phys. Rev., D28, 2363-72.
Gallavotti, G. ( 1983). The Elements of Mechanics. Springer : New York.
Goenner, H . ( 1970). J. Math . Phys., 1 1 , 1645 .
Grishchuk, L . P . and Kopejkin, S . M . ( 1983). Pis'ma Astron. Zh ., 9, 436--40 (Sov. Astron.
Lett., 9, 230-2).
Grishchuk, L . P. and Kopejkin , S. M . ( 1986) . In Relativity in Celestial Mechanics and
Astrometry (IAU Symposium 1 14, Leningrad , 1985), ed . J. Kovalevsky and V. A .
Brumberg , p p . 19-34. Reidel : Dordrecht.
Giirses, M. and Walker, M. ( 1984). Phys. Lett., 101A, 15-19.
Hagihara, Y . ( 1972). Celestial Mechanics. M IT Press : Cambridge, M ass.
Havas, P . ( 1957). Phys. Rev., 108, 135 1 .
Havas, P . ( 1 979) . I n Isolated Gravitating Systems in General Relativity, ed. J . Ehlers,
pp. 74- 1 5 5 . North-Holland : Amsterdam.
Havas, P. and Goldberg, J. N. ( 1962). Phys. Rev., 128, 398--4 14.
Hellings, R. W . ( 1986). Astron. J., 91 , 650-9 .
Hogan, P. A . and McCrea, J . D . ( 1974) . Gen. Rel. Grav. , 5, 79- 1 13.
Hulse, R . A . and Taylor, J . M . ( 1975). Astroph;!s. J. (Lett.), 195, L5 1-53 .
Ibanez, J . and M artin, J . ( 1982). Gen. Rel. Grav., 14, 439-5 1, and Phys. Rev., D26,
384-9.
Infeld, L . and Plebanski, J . ( 1 960) . Motion and Relativity. Pergamon Press : Oxford .
Isaacson, R . A . , Welling, J . S. and Winicour, J . ( 1984) . Phys. Rev. Lett., 53, 1870-2.
Isaacson, R. A. and Winicour, J. ( 1968). Phys. Rev., 168, 145 1-6.
Jardetzky, W. S. ( 1958). Theories of Figures of Celestial Bodies. Interscience : New York.
Kates, R. E. ( 1980a). Phys. Rev., D22, 1853-70.
196 T. Damour
Kates, R. E. ( 1980b) . Phys. Rev., 022, 187 1-8.

Kates, R . E. ( 198 1). Ann. Phys. (NY), 132, 1- 17.
Kerlick, G . D. ( 1980). Gen. Rel. Grav., 12, 467, 52 1 .
Kerr, R . P . ( 1959). Nuov. Cim., 13, 469-9 1 , 492-502, 673-89.
Kopejkin, S. M . ( 1985). Astron. Zh., 62, 889-904 (in Russian).
Kovalevsky, J. and Brumberg, V. A., eds. ( 1986). Relativity in Celestial Mechanics and
Astrometry (IAU Symposium 1 14, Leningrad, 1985). Reidel : Dordrecht .
Kuhnel, A . ( 1963). Acta Phys. Polon., 24, 399.
Kuhnel, A. ( 1964). Ann, Phys. (NY), 28, 1 16-33.
Lagerstrom, P . A., Howard, L . N . and Liu, C. S . ( 1967). Fluid Mechanics and Singular
Perturbations: A Collection of Papers by Saul Kaplun . Academic Press : New York .
Laplace, P . S . ( 1 805). Traite de Mecanique Celeste, vol. 4, pt 2, Book X , chapt. VII.
Courcier : Paris, an XIII (reprinted in Oeuvres completes de Laplace, vol . 4, pp. 3 1 4-27.
Gauthier-Villars : Paris, 1880).
Leipold, G. and Walker, M. ( 1977). Ann. Inst. H. Poincare A (Phys. Theor.), 27, 6 1-7 1 .
Lestrade, J. F., Chapront, J. and Chapront-Touze, M . ( 1982). I n High-Precision Earth
Rotation and Earth-Moon Dynamics, ed . 0. Calame, pp. 2 1 7-24. Reidel : Dordrecht.
Levi-Civita, T. ( 1937a). Am . J. Math., 59, 9-22.
Levi-Civita, T. ( 1 937b). Am. J. Math., 59, 225-34.
Levi-Civita, T. ( 1950). 'Le probleme des n corps en relativite generale, Memorial des
Sciences Mathematiques, 1 16. Gauthier-Villars : Paris.
Lichnerowicz, A. ( 1955). Theories Relativistes de la Gravitation et de l'Electromagnetisme.
Masson : Paris.
Lorentz, H . A. and Droste, J. ( 19 17). Versl. K . Akad. Wet. Amsterdam, 26, 392 (part I)
and 649 (part II). English translation in H. A. Lorentz, Collected Papers, ed .
P. Zeeman and A. D. Fokker, vol. V, pp. 330-55. Martinus Nijhoff: The Hague, 1937.
McCrea, J. D. and O 'Brien, G. ( 1978). Gen. Rel. Grav. , 9 , 1 10 1-18.
Manasse, F. K . ( 1963). J. Math. Phys., 4, 746-6 1 .
Martin, C . F., Torrence, M . H . and Misner, C . W . ( 1985). J . of Geophys. Res., 90,
9403- 10.
Mashhoon, B. ( 1 984). Gen. Rel. Grav., 16, 3 1 1 .
Mashhoon, B. ( 1985). Found. Phys., 1 5, 497.
Misner, C. W . , Thorne, K. S. and Wheeler, J. A. ( 1973). Gravitation. Freeman : San
Francisco (the many-body system is treated in Exercise 39 . 1 5, pp. 109 1-5) .
Newton, I . ( 1947). The Mathematical Principles of Natural Philosophy, ed . F. Cajori .
University of California Press : Berkeley.
Nordtvedt, K., Jr ( 1968). Phys. Rev., 1 70, 1 186-7.
Ohta , T., Okamura, H . , Kimura, T. and Hiida, K. ( 1973). Prog . Theor. Phys., 50,
492-5 14.
Ohta, T., Okamura, H., Kimura, T. and H iida, K. ( 1 974a). Prog. Theor. Phys., 5 1 ,
1220-38.
Ohta, T., Okamura , H . , Kimura, T. and H iida , K. ( 1974b). Prog. Theor. Phys., 5 1 ,
1 598- 16 12.
Papapetrou , A . ( 195 1). Proc. Phys. Soc., A64 , 57-75.
Peres, A . ( 1959). Nuov. Cim. , 1 1 , 6 1 7 , 644 ; 13, 437.
Peres, A. ( 1960) . Nuov. Cim., 15, 35 1-69.
Persides, S. ( 197 1a). J. Math. Phys., 12, 2355-6 1 .
Persides, S . ( 197 1b). Astrophys. J., 1 70, 479-98.
Peters, P. C . ( 1 964). Phys. Rev., 136, B 1224-32 .
Peters, P. C. and Mathews, J . ( 1963). Phys. Rev., 131 , 435-40.

Poincare, H . ( 1 934) . Science et Methode, Book 1, chapt. II, p. 34 . Flammarion : Paris.
Poincare, H . ( 1957). Les methodes nouvelles de la mecanique celeste. Dover : New York
(reprint of the first edition by Gauthier-Villars, Paris, 1892-99) .
Riesz, M . ( 1949). Acta Mathematica , 8 1 , 1-223.
Selia.fer, G. ( 1983). Mitt. Astron. Ges . , 58, 135-7.
Schafer, G . ( 1984). Phys. Lett., lOOA, 128-9.
Schafer, G . ( 1985). Ann . Phys. (NY), 161 , 8 1- 100.
Schafer, G . ( 1986) . Gen. Rel. Grav., 18, 255-70.
Schattner, R. ( 1978). Gen. Rel. Grav., 10, 377-93, 395-9.
Schutz, B. F. ( 1980). Phys. Rev., 022, 249-59.
Schutz, B. F . ( 1984). In Relativistic Astrophysics and Cosmology, ed . X. Fustero and
E .Verdaguer, pp. 35-97 . World Scientific : Singapore.
Schutz, B. F. ( 1985). In Relativity, Supersymmetry and Cosmology, ed . 0. B ressan et al.
(SILARG V, Bariloche, 1985), pp. 3-80. World Scientific : Singapore.
Spyrou , N. ( 1977). Gen . Rel. Grav. 8, 463-89 (this work has been extended by N. Spyrou
( 1977-85) in a series of papers in the same journal).
Spyrou, N . ( 1978). Cel. Mech . , 18, 3 5 1-70.
Stephani, H . ( 1964) . Acta Phys. Polan. , 26, 1045-60.
Synge, J. L. ( 1960). Relativity: The General Theory. North-Holland : Amsterdam.
Synge, J. L. ( 1 969) . Proc. Roy. Ir. Acad. , A67 , 47 .
Synge, J . L . ( 1 970). Proc. Roy. Ir. Acad., A69, 1 1 .
Taylor, J . H . , Fowler, L . A . and McCulloch, P . M . ( 1979). Nature, 277, 437-40.
Taylor, J. H . and Weisberg, J. M . ( 1982). Astrophys. J., 253, 908-20.
Thome, K. S . ( 1969). Astrophys. J., 158, 997- 1019.
Thorne, K. S. and Kovacs, S. J . ( 1975). Astrophys. J., 200, 245-62 (see also two
subsequent papers by S. J. Kovacs and K. S. Thorne ( 1977, 1978) in the same
journal).
Thorne, K . S . and H artle, J . B . ( 198 5). Phys. Rev., D31 , 18 15-37.
Tisserand , F . ( 1960). Traite de Mecanique Celeste, vol . I . Gauthier-Villars : Paris (reprint
of the first edition of 1889).
Van Dyke, M . ( 1975) . Perturbation Methods in Fluid Mechanics. Parabolic Press :
Stanford , California (annotated edition).
Wagoner, R. V. ( 1975) . Astrophys. J., 196, L63-5 .
Walker, M . ( 1984). In General Relativity and Gravitation, ed . B . Bertotti et al. (GR 10,
Padua, 1983), pp . 107-37. Reidel : Dordrecht .
Walker, M . and Will , C. M . ( 1980). Astrophys. J. (Lett.), 242, L 129-33.
Weisberg, J . M. and Taylor, J . H. ( 1 984) . Phys. Rev. Lett. , 52, 1 348-50.
Westpfahl , K. ( 1985). Fortschritte der Physik, 33, 4 17-93.
Westpfahl , K . and Goller, M. ( 1979) . Lett. Nuov. Cim. , 26, 573-6.
Westpfahl, K . and Goller, M . ( 1980). Lett. Nuov. Cim., 27, 16 1 .
Westpfahl, K . and Hoyler, H . ( 1980). Lett. Nuov. Cim., 27, 58 1 .
Will , C. M . ( 1974) . I n Experimental Gravitation, ed . B . Bertotti (56th E . Fermi School,
Varenna, 1972), pp. 1- 1 10. Academic Press : New York (see pp. 64--- 5) .
Will, C. M . ( 1979) . In General Relativity: An Einstein Centenary Survey, ed . S . W .
Hawking and W. Israel , p p . 24---8 9. Cambridge University Press : Cambridge.
Will, C. M . ( 198 1). Theory and Experiment in Gravitational Physics, Cambridge
Will, C. M . ( 1986). Can. J. Phys., 64, 140-5 .
198 T. Damour
Will, C. M . and Eardley, D. M. ( 1 977). Astrophys. J., 212, L9 1-4.

Winicour, J. ( 1 983). J. Math . Phys. , 24, 1 1 93-8 .
York, J. W. Jr ( 1 983). In Gravitational Radiation , ed . N. Deruelle and T . Piran ,
pp. 175-20 1 . North-Holland : Amsterdam .
7
Dark stars: the evolution of an idea
W E RN E R I S R A E L t
7. 1 Introduction
Theory may often delay understanding of new phenomena observed
with new technology unless theorists are quite open-minded as to what
types of physical laws may need to be applied : conservatism is unsafe.
. . . In astrophysics, historically, theories have only seldom had
predictive usefulness as guides to experimenters.
(Greenstein, 1984)
How difficult were these advances? If we now acquaint ourselves with
some of the detours, errors and psychological blocks, it is not to console
ourselves with the thought that even great physicists do not always
move in a straight line, but rather to grasp in some measure how
difficult it really was.
(Hund, 197 1)
To pause halfway up a steep hill , look down at one's starting-point barely
half a mile below and recall the arduous, winding path one has actually
followed , is a salutary exercise in humility. The subject of this essay - the
evolution of our ideas since the age of Laplace concerning the dark objects
that populate the universe - is now at that middle distance where a
retrospective look begins to discern broad contours without yet losing all of
the detail that will in time become laundered and streamlined by Darwinian
selection of citations into the handful of names and dates of the potted
histories.
Tracing this evolution, with its meanderings and vicissitudes, as a
continuous process, means abandoning attempts at a 'balanced' account
which gives consideration to ideas in proportion to their evolutionary
t Work supported by Canadian Institute for Advanced Research and by Natural
Sciences and Engineering Research Council of Canada.
200 W Israel
fitness. Some dinosaurs have to be resurrected. The old and discarded is

often scaffolding for the new : Faraday's concept of field grew out of the
aether, Carnot's thermodynamics from the notion of a caloric fluid.
Conversely, the attempt to trace the broad current of ideas and opinions
without getting mired in technicalities , has meant that many technical
advances that have been central to the development are mentioned only in
passing, or not at all ,just because they were neither controversial nor wrong
in an interesting way.
As basic science moves further from everyday reality, a number of
questions increasingly solicit attention . How does the collective scientific
mind evolve toward new concepts? What keeps the scientific wagon on the
right track? With science increasingly under the sway of fads, how far dare
one trust scientific consensus as an autostabilizing mechanism? The
evolutionary picture suggested by the present account is not altogether
reassuring on this score. The black hole idea is more than 200 years old :
general relativity more than 70. Yet, if our present ideas about gravitational
collapse and the existence of neutron stars and black holes happen to be
near the truth, a scarce 25 years ago virtually the entire scientific community
was complacently wandering in the darkness. It is not easy to detect in this
story anything that often resembles the Kuhnian cycle of paradigm and
revolution. Rather, one sees a meandering path that resolves into a
Brownian motion at the microlevel of individual scientists. Perhaps it is not
altogether flippant to suggest that the evolutionary picture that naturally
presents itself is of a substantially classical motion (but with considerable
quantum spread) that arises by constructive interference from a probing of
all possible paths, very occasionally interrupted by barrier tunneling at
places where the incident current becomes sufficiently large. Whether this
motion will ultimately approach a fixed point or limit cycle, and whether
there is more than one, are questions for future generations.
This narrative comes to an end around 1974, when the main pieces of the
pattern, as we now construe it, were in place. As it moves nearer to the
present, the perspective inevitably becomes more subjective. Others, located
at diverse research centres around the globe, will have experien�ed the
events recounted here in their own very different ways. While I have striven to
keep technicalities to a minimum, this is basically a historical sketch and not
an exposition of the subject itself. For the newcomer, Shklovsky ( 1978) on
stellar physics, and Israel ( 1983), on black holes, are recent semipopular
expositions with extensive references to collateral reading.
The many people who patiently endured my questions and who are not to
Dark stars: the evolution of an idea 20 1
blame for my misconstruction of their replies or the way it has all been
pieced together, include J. D . Barrow, S. Chandrasekhar, D . Finkelstein,
S. W. Hawking , D. P. Hube, M . D. Kruskal , W. H. McCrea, M. J. Rees,
K. S. Thorne and G . M . Volkoff, and I should like to extend. to them my
deepest thanks. My thanks are also due to Professor Z. Maki and Professor
K. Nishijima, the successive Directors of the Research Institute for
Fundamental Physics, Kyoto University, and their colleagues, for their kind
hospitality at Yukawa Hall , where this work was begun. I am indebted to
the staff of the Scientific Periodicals Library, Cambridge University, for
making their unique resources available, and uncomplainingly bringing out
an endless stream of volumes from their stacks, and to Yoshiko Fujinaka,
Mary Yiu and Cheryl Torbett for their superb technical assistance.
7.2 Early speculations (1 784-192 1)
Unseen bodies may, for aught we can tell , predominate in mass over the
sum-total of those that shine : they supply possibly the chief part of the
motive power of the universe.
(Clerke, 1 903)
In regions where our ignorance is great, occasional guesses are
permissible.
(Lodge, 192 1)
The name of the Reverend John Michell (MA (Cantab .) 1 752, BD 1 76 1 ) fell
so quickly into obscurity after his death that two of the English scientific
classics of our century, Eddington's Internal Constitution of the Stars ( 1 926)
and Hawking and Ellis's Large Scale Structure of Space-Time ( 1973) give
credit to a Frenchman , P. S. Laplace ( 1 796) , as having been first to suggest
that 'the attractive force of a heavenly body could be so large that light could
not flow out of it'.
Yet in his day M ichell's reputation as a natural philosopher stood second
in English circles only to that of his contemporary and friend, Henry
Cavendish. The paper communicated to the Royal Society by Cavendish on
November 27, 1783, in which Michell expounded this and many related
ideas, caused such a stir in London circles that it overshadowed exciting
news coming from Paris about Coulomb's electrical experiments,t as extant
t Coulomb's invention of the torsion balance had actually been anticipated by M ichell,
and the inverse-square law by Cavendish, using the now celebrated null method of
measuring the force inside a hollow charged conductor. Both Cavendish and Michell
were reclusive personalities who published very little of their wide-ranging work . Many
of Cavendish's electrical researches were found after his death in a number of sealed
packets of papers, eventually published, under the editorship of Maxwell, in 1879. None
of M ichell's unpublished work survives.
202 W Israel
correspondence of Sir Joseph Banks (then President of the Royal Society)

bears witness.
Michell's ( 1784) paper is far from airy speculation, but a concrete
proposal , worked out in detail , for obtaining information about stellar
distances, masses and sizes by observing the retardation of light emitted
from their surfaces.t This was at a time when , despite efforts extending over
a century, no semi-annual stellar parallax had yet been found , and some
astronomers doubted whether they would ever be detectable.
Michell's work is of course based on two of Newton's major hypotheses,
universal gravitation and the corpuscular theory of light. He says :
Let us now suppose the particles of light to be attracted in the same
manner as all other bodies with which we are acquainted ; . . .
gravitation being , as far as we know, or have any reason to believe, a
universal law of nature.
Even Michell's geometrical style of reasoning is quintessentially Newtonian .
Russell McCormmach (1968) , in his moving account of the friendship and
collaboration of M ichell and Cavendish, describes their deep attachment to
the Newtonian scheme of things :
Newton wrote in the Principia that all bodies are to be regarded as
subject to the principle of gravitation . Every body, bowever great or
small, is related to every other body in the universe by a mutual
attraction . It was this postulated universality of the force of gravit y
which contributed so greatly to the order and unity of the Newtonian
world . This unity was, for its followers, an untested article of faith for
nearly a century after the Principia. During this time the evidence of
gravitational attraction continued to be drawn from the motions of the
earth, moon , planets, comets, and falling bodies - phenomena which
span an intermediate range of masses , sizes and distances . In three
domains of experience, involving the extreme upper and lower limits of
masses and dimensions, the action of gravity had not yet been
observed ; the gravity of the 'fixed' stars ; the mutual attraction of
terrestrial bodies ; and the gravitation of light. The task of deducing
observable consequences from each of these supposed instances of
universal gravitation fell to the Reverend John M ichell ( 1724- 1793), a
teacher at Cambridge ( 1749- 1763) and afterwards Rector of Thornhill ,
Yorkshire. H is renowned London friend Henry Cavendish ( 173 1- 18 10)
encouraged him in these researches and became involved in the
t Michell calculated that a retardation of 5 per cent would change the refractive index
sufficiently to make it easily measurable with a prism.
Dark stars: the evolution of an idea 203
resulting observational and experimental questions. The immersion of

Michell and Cavendish in gravitational studies was an essential feature
of their commitment to a unified Newtonian world. Their commitment
had yet a broader significance : Newton's theory of gravitation inspired
their image of physical reality, and it served as their model of an exact
science . . . .
The freshness and scope of Michell's scientific imagination help to
explain Cavendish's unique interest in their correspondence. But there
is more to it than this. Michell's isolation was not wholly physical ; it
was intellectual too, and Cavendish shared that isolation ; their true
bond was philosophical . They believed with Newton that the purpose
of natural philosophy is to discover the mathematical laws of the
attracting and repelling forces between the particles of bodies, and to
deduce new phenomena from the forces. They stood out against the
current of British natural philosophy, which by then was neither
mathematical nor primarily concerned with laws of force. The best
science of the day was one-sidedly experimental ; and the speculative
interest was generally focused on the ether, and on force-denying
mechanisms, and on the physical connections between the post
Newtonian weightless fluids.
While observational questions are always paramount in Michell's paper,
he also contemplated the possibility that there might exist stars much larger
than the sun for which the velocity of escape exceeds the speed of light. Even
here, he has a prescient suggestion of pow such stars might be detectable :
I f there should really exist i n nature any bodies whose density is not less
than that of the sun, and whose diameters are more than 500 times the
diameter of the sun , since their light could not arrive at us . . . we could
have no information from sight ; yet, if any other luminous bodies
should happen to revolve about them we might still perhaps from the
motions of these revolving bodies infer the existence of the central ones
with some degree of probability, as this might afford a clue to some of
the apparent irregularities of the revolving bodies . . . ; but as the
consequences of such a supposition are very obvious, and the
consideration of them somewhat beside my present purpose, I shall not
prosecute them any further.
A remarkable development of this line of thought occurs in the work of
the German astronomer Johann Georg von Soldner ( 1 80 1 ; see Jaki, 1978).
Soldner calculated the Newtonian deflection oflight corpuscles passing near
a star. He also speculated on the possibility that the stars of the Milky Way
204 W. Israel
might be orbiting a central, very massive dark object of the sort postulated
by Laplace, but decided against it on the grounds that the resulting proper
motions would be too large not to be noticed.
Speculation about invisible stars went quickly out of favour as the wave
theory of light gained ascendancy especially after the discovery of
interference by Thomas Young in 180 1 , and removed any reason to believe
that light should be affected by gravity. References to this hypothesis were
deleted from the 1808 and later editions of Laplace's Exposition du
Systeme du Monde . But it was natural that it would resurface, albeit in a
significantly different form, after the success of the 1 9 19 eclipse expedition.
In February 1920, A. Anderson of University College, Galway, made an
intriguing speculation in the Philosophical Magazine :
We may remark, though perhaps the assumption is very violent, that if
the mass of the sun were concentrated in a sphere of diameter 1 .47
kilometres, the index of refraction near it would become infinitely great ,
and we should have a very powerful condensing lens, too powerful
indeed, for the light emitted by the sun itself would have no velocity at
its surface. Thus if, in accordance with the suggestion of Hemholtz, the
body of the sun should go on contracting there will come a time when it
is shrouded in darkness, not because it has no light to emit, but because
its gravitational field will become impermeable to light.
This was an extraordinary anticipation of the gravitational collapse
scenario which was to be so adamantly resisted in the 1930s and for a
quarter of a century afterward .
The convenient fiction , due to Eddington, that the paths of light rays in a
gravitational field could be formally described and visualized as
propagation in an 'aether' of variable refractive index , was imaginatively
taken up by Sir Oliver Lodge, whose firm belief in the aether had somehow
withstood the assault of the 1905 relativistic revolution (see, e.g. his
contributions to the discussion of the results of the 1 9 19 Sobral expedition ,
Lodge, 1920). In a remarkable address to a Student's Science Club at the
University of Birmingham in 192 1 , he presented on this basis an essentially
correct physical picture of the full range of black holes that are of
astrophysical interest today :
If light is subject to gravity, if in any real sense light has weight , it is
natural to trace the consequences of such a fact. One of these
consequences would be that a sufficiently massive and concentrated
body would be able to retain light and prevent its escaping . And the
body need not be a single mass or sun , it might be a stellar system of
exceedingly porous character so that light could penetrate freely into

the interior . . . .
. . . we find that a system able to control and retain its light i;nust have
a density and size comparable to
pR 2
= · · · =
1 .6 x 10 2 7 c.g .s.
It is hardly feasible for any single mass to satisfy this condition ; either
the density or the size is too enormous . . .
For a body of density 101 2 , - which must be the maximum possible
density , as its particles would be then all jammed together, - the radius
need only be 400 kilometres . . . .
If a mass like that of the sun (2.2 x 103 3 grammes) could be
concentrated into a globe of about 3 kilometres in radius, such a globe
would have the properties above referred to ; but a concentration to
that extent is beyond the range of rational attention . . . .
But a stellar system - say a super spiral nebula - of aggregate mass
equal to 101 5 suns . . . might have a group radius of 300 parsecs . . . with
a corresponding average density of 10 - 1 5 c.g.s. without much light
being able to escape from it. This really does not seem an utterly
impossible concentration of matter.
(Lodge, 192 1)
At a time when astronomers still believed, in spite of their awareness that
stellar material was highly ionized, that this material could not be
compressed to a density much higher than that of water (e.g. Eddington ,
192 1), Lodge's casual acceptance of the possibility of attaining nuclear
density is truly noteworthy. This imaginative contribution was yet another
stillbirth , and it was to be not the last before the black hole idea finally came
into its own more than 40 years later.
7.3 White dwarfs : the first compact massive objects (1910-26)

The message of the companion of Sirius when it was decoded ran : 'I am
composed of material 3000 times denser than anything you have ever
come across ; a ton of my material would be a little nugget that you
could put in a matchbox' . What reply can one make to such a message?
The reply which most of us made in 1 9 14 was - 'Shut up. Don't talk
nonsense'.
(Eddington, 1927)
The first astronomical bete noire of our century, the white dwarfs, cast their
shadow on a generation still grounded in a Victorian sense of the
206 W Israel
permanence of things and the Lucretian doctrine of atoms that could 'not be
swamped by any force, for they are preserved indefinitely by their absolute
solidity'. As the web of observation and theory slowly tightened, the
scientific reaction - first disregard, then dismay, yielding only gradually to
the beginnings of acceptance - set a pattern for the disclosures yet to come.
The companions of Procyon and Sirius had puzzled astronomers since
before the turn of the century. Comparable in mass with the sun , they were
more than a hundred times dimmer, 'a condition that we are powerless to
explain' (Campbell , 1 9 1 3). Superficially, there seemed to be two
alternatives :
Either they have a far less surface brilliancy than the sun or their density
is much greater. There can be no doubt that the former is the case.
(Newcomb , 1908)
This emphatic conclusion would have caused no misgivings if these
objects had been unmistakably red and thus of low surface temperature :
their dimness would have been explicable and not at all unusual . But, as far
as it could be distinguished from the overwhelming brilliance of Sirius itself,
the companion appeared nearly as white to the eye. t In December 1 9 1 5 ,
Walter S. Adams announced that he had finally succeeded in securing a
spectrogram with the Cassegrain reflector at Mt Wilson as the companion
passed to its furthest distance from the primary in its 49-year orbit. It
showed a spectrum identical with that of Sirius. Puzzlement now gave way
to some embarrassment , since spectral intensity measurements of 109 stars
by Wilsing and Scheiner at Potsdam during the previous five years had
demonstrated (as had earlier theoretical work by Schwarzschild) that to a
fair approximation stars radiate like black bodies, with effective
temperatures well correlated with colour index and spectral type.t If the
observations were taken at face value, the dimness of Sirius B could not be
attributed to a low surface temperature. Presumably the reaction of the
anonymous reporter for The Observatory ( 1 9 16) was not untypical : 'The
results rather suggest that the spectrum obtained is due to light from Sirius
t In the case of Procyon, the spectral class of the companion remains somewhat
uncertain to this day due to the proximity of the bright component (Liebert, 1980).
t According to the classification system perfected at the Harvard College Observatory in
the early 1 900s, stellar spectra may be divided into several reasonably well-defined types,
labelled 0, B, A, F, G, K, M in order of decreasing surface temperature. Sub-classes a re
denoted by the digits 0 to 9, so that AO is the hottest star of type A . The colour of a star
is a clue to its spectral type. Stars of types 0 and B (characterized by prominence of
helium lines) are bluish-white, type A (hydrogen lines) white, types F and G (calcium
lines) yellow, types K and M (metallic lines) orange to red .
Dark stars: the evolution of an idea 20�
reflected by the companion' . Adams had already anticipated the objection :

It is . . . by no means necessary to have recourse to this explanation,
since in the case of o 2 Eridani, where there can be no question of
reflected light, we know of a similar case of a star of very low intrinsic
brightness which has a spectrum of type A0.
(Adams, 19 15)
Some months earlier, the Danish astronomer Ejnar Hertzprung (19 1 5)
had taken note of this system as the single exception known to him to the
general rule that 'stars between absolute magnitudes + 3 and + 8 were
approximately all of the same colour', and commented , 'This exception is, in
fact , very strange' .
The anomaly o f o 2 Eridani had been noticed b y others even before this.
Years later, Henry Norris Russell of Princeton was to recall a day in 1 9 10
when the first white dwarf made itself known to an inner circle of three :
I was visiting my friend and generous benefactor, Prof. Edward C.
Pickering [director of the Harvard College Observatory ( 1876-19 19)] .
With characteristic kindness he had volunteered to have the spectra
observed for all the stars - including comparison stars - which had been
observed in the observations for stellar parallax which Hinks and I
made at Cambridge, and I discussed. This piece of apparently routine
work proved very fruitful - it led to the discovery that all the stars of
very faint absolute magnitude were of spectral type M . In conversation
on this subject (as I recall it), I asked Pickering about certain other faint
stars, not on my list, mentioning in particular 40 Eridani B .
Characteristically, h e sent a note- t o the Observatory office and before
long the answer came (I think from M rs Fleming) that the spectrum of
this star was A. I knew enough about it, even in those paleozoic days, to
realize at once that there was an extreme inconsistency between what
we would then have called 'possible' values of the surface brightness
and density. I must have shown that I was not only puzzled but
crestfallen, at this exception to what looked like a very pretty rule of
stellar characteristics; but Pickering smiled upon me and said : 'It is just
these exceptions that lead to an advance in our knowledge', and so the
white dwarfs entered the realm of study.
(Russell, 1939, 1944)
It was, however, left to a 23-year-old Estonian astrophysicist, Ernest J.
Opik, then working in Moscow, to take upon himself the odium of
confronting the paradox openly. In May 1916 he submitted a paper in which
he calculated mean densities for 40 stars in 39 binary systems whose spectral
208 W. Israel
class was known at the time. (With communication impeded by the war,
Opik as yet was unaware of Adams's work on the spectrum of Sirius B .) He
wrote :
Among the binaries with known orbital elements there is one which is
not included in the preceding discussion, O z Eridani ; . . . the density
according to equation (6) . . . would be 25 000. This impossible result
indicates that in this case our assumptions are wrong ; the only possible
explanation is that, however high the temperature, the surface
brightness or the radiating power is very low ; probably Oz Eridani is a
pair of very rarefied nebulae.
(Opik , 1 9 1 6)
The key to the true explanation was forged just six months later and in a
closely related context, but it took another eight years before anyone saw the
connection. On 8 December 19 16 Eddington presented to the Royal
Astronomical Society the first of his pioneering papers on radiative
transport in stellar interiors. In the ensuing discussion, Jeans pointed out
that the value µ = 54 which Eddington had adopted for the mean molecular
weight of the stellar material was almost certainly an overestimate, since 'for
these temperatures and energy we have very hard Rontgen radiation , and so
the atoms in the gas will be smashed up' (The Observatory, 19 17). Jeans
( 1928) later traced this idea back to Descartes ( 1644), who had conjectured
that the sun and fixed stars were made of matter 'which possesses such
violence of agitation that , impinging upon other bodies, it gets divided into
indefinitely minute particles'. Under such conditions, Jeans was now
pointing out, electrons normally bound to much heavier nuclei would be set
free as independent particles, thus greatly reducing the mean molecular
weight of the material.
In his subsequent work, Eddington ( 19 1 8 , 192 1 ) revised his estimate for µ
downwards to values in the range 2 to 4, but, like everyone else (e.g. Jeffreys,
19 18), he still believed that, in stars like the sun , material would deviate
drastically from a perfect gas. It was more than just 'common sense' that
deterred the astronomers from questioning this axiom. Breakdown of the
gas laws at ordinary densities was a pillar of the giant-and-dwarf
evolutionary theory of Russell (19 14), which held sway well into the 1920s.
In Russell's version of the theory·, stars were pictured as beginning their lives
as gaseous red giants, contracting and heating up as they (supposedly)
evolve towards the blue end of the main sequence, at which point their
material becomes liquid and begins to cool. The stars then evolve along the
main sequence to end their luminous phase as red dwarfs. The discovery of
the mass-luminosity relationship by Eddington ( 1924a) cut the ground from

under this theory, since it showed that different locations on the main
sequence correspond to different masses. It had an even more far-reaching
and remarkable result :
We must recall that the theory was developed for stars in the condition
of a perfect gas . . . . But in the left half of the diagram we have the sun
whose material is denser than water, Krueger 60 denser than iron . . . .
What business have they on the curve reserved for a perfect gas? . . . The
shock was even greater than I can well indicate to you, because the
great drop in brightness when the star is too dense to behave as a perfect
gas was a fundamental tenet in our conception of stellar evolution. On
the strength of it the stars had been divided into two groups known as
giants and dwarfs, the former being the gaseous stars and the latter the
dense stars.
(Eddington , 1927)
Eddington communicated these results to a memorable meeting of the
Royal Astronomical Society (RAS) on 14 March 1924. Even now, he did not
yet feel free to take it quite for granted that ionization would reduce atomic
sizes :
Does losing an electron mean that the boundary of the atom is gone? Is
the boundary or merely the boundary-stone removed? I think that
most physicists would consider that the removal of the outer electrons
means the removal of the boundary.
(Eddington , 1924a)
Today such hesitancy makes curious reading until one recalls that
quantum mechanics was still two years in the future, and Bohr's ad hoc
quantization rules for the placement of electronic orbits offered no clue to a
dynamical explanation for the size and rigidity of atoms. Eddington now
proceeded to the climax of his address :
There is an interesting class of stars known as the 'white dwarfs', of
which the companion to Sirius is a famous example. It presents a
cu rious puzzle . . . . If . . . it has the ordinary radiative power associated
with F-type stars, we can work out its surface area and its diameter,
which comes out not very much larger than the earth ; its mass is about 1
that of the sun , so that the density would be enormous - about
50 000 gm per c.c. That argument is well-known, but I think most of us
have mentally added 'which is absurd'. According to the present
conclusions, however, it is not absurd ; so that, unless the spectroscopic
classification has deceived us, the companion of Sirius may be an actual
210 W. Israel
example of how disrupted atoms may pack together to a far higher

density than ordinary matter.
(Eddington, 1924a)
Eddington felt sufficiently bold not to wait for his colleagues' reactions to
this 'fascinating train of ideas' (as Jeans was to phrase it) before following up
the consequences. About six weeks earlier, he had written to Walter Adams
at Mt Wilson :
I have lately been wondering if you would find it possible with your
great instrument to measure the radial velocity of the companion of
Sirius - with a view to determining its density by means of the Einstein
shift of the spectral lines (compared with Sirius).
I take it that it is about 10m fainter than Sirius and as it is of the same
spectral type it should have a surface 10 000 times smaller and a radius
100 times less . . . . Of course this involves a density of about 100 000
times the density of water, which seems incredible. But I have recently
been entertaining the wild idea that it may be just possible . . . .
I should scarcely venture to suggest your following this wild idea if I
did not regard a negative result as also of interest . . . a very definite
challenge to thermodynamicists . . . .
(Eddington, l 924b)
Adams replied on 12 February, promising to attempt measurements, and
added :
I had realised the shift would be large but had not attempted to
compute it . . . I have looked at these plates many times with a view to
.
m easurement but have always laid them aside . . . .

On 2 March he wrote again :
We have obtained some results on the companion of Sirius which I
believe will interest you . . . . The simple mean is + 20 km per sec ; or if
the best line is given double weight the mean is + 23 km . . . . I should
very much appreciate your opinion on these results. The probable error
is, of course, large . . . .
Eddington's initial response on 22 March was cautious :
I feel pretty well convinced that large densities are possible, but whether
we have been lucky enough to find an actual specimen is another
question. It seems almost too good to be true. It looks all right , but
owing to the difficulty of the measurement and the rather large
probable error I regard your result with some caution - as I think you
would wish, and shall not say anything about it publicly at present. I do
not know whether you are proposing to take further spectrograms or
measures before you publish it. It is evidently of exceptional theoretical

importance besides being a great practical triumph over difficulties.
Fifteen months later Adams ( 1925) was ready to publish his conclusions
and Eddington's satisfaction was unalloyed :
I am very delighted that your measures have turned out so well and am
deeply interested in the details of the paper. But it is very staggering that
it should turn out like this . . . .
(Eddington , 1925)t
Eddington's classic, The Internal Constitution of the Stars, was a survey of
the subject as it stood just prior to the discovery of wave mechanics.
Discussing the white dwarfs he poses, in his characteristic Carrollian
fashion, a conundrum which has come to be known as Eddington 's
paradox :
I do not see how a star which has once got into this compressed
condition is ever going to get out of it . So far as we know, the close
packing of matter is only possible so long as the temperature is great
enough to ionise the material . When the star cools down and regains
the normal density associated with solids, it must expand and do work
against gravity . The star will need energy in order to cool . . . . We can
scarcely credit the star with sufficient foresight to retain more than 90
per cent [of its internal energyJ in reserve for the difficulty a waiting it. It
would seem that the star will be in an awkward predicament when its
supply of sub-atomic energy ultimately fails. Imagine a body
continually losing heat but with insufficient energy to grow cold !
(Eddington, 1926)
This dilemma, in one form or another, would dog the theorists for more
than three decades. In 1926, however, it appeared that deliverance was at
hand . R . H . Fowler pointed out that, on the basis of the new quantum
statistics, even an absolutely cold assembly of electrons confined to a finite
volume must retain a finite spread of momenta, and hence a pressure, since
the Pauli exclusion principle would not allow more than two electrons (with
opposite spins) to occupy the same quantum phase cell. Milne ( 1930)
showed that this zero-point pressure can balance a cold star against gravit y
at a uniquely determined radius (a decreasing function of mass) that
corresponds well with the actual sizes of the white dwarfs. Fowler gave a
graphic description of the condition of matter in such a star :
t Since the 1950s it has been generally admitted that both the theoretical and observed
redshifts for Sirius B are highly uncertain. However, there is reasonably good agreement
in the case of o 2 Eridani.
212 W Israel
The black dwarf material is best likened to a single gigantic molecule in

its lowest quantum state. On the Fermi-Dirac statistics its high density
can be achieved in one way only, in virtue of a corresponding great
energy content. But this energy can no more be expended in radiation
than the energy of a normal atom or molecule. The only difference
between black dwarf matter and a normal molecule is that the molecule
can exist in a free state while black dwarf matter can only so exist under
very high external pressure.
(Fowler, 1929)
It thus seemed for a while, at the end of the 1920s, that one of the most
bizarre chapters in the history of astrophysics had been brought to a
rounded and satisfying conclusion. However, even before the decade was
quite out, the demon exorcised by Fowler had covertly wormed its way
back.
7.4 The Chandrasekhar limit (1929-35)
The ground will fall away from our feet, its particles dissolved amid the
mingled wreckage of heaven and earth. The whole world will vanish
into the abyss, and in the twinkling of an eye no remnant will be left but
empty space and invisible atoms. At whatever point you allow matter
to fall short, this will be the gateway to perdition.
(Lucretius, 55 BC )
Edmund C. Stoner of Leeds University is best known for his pioneering
work on the application of Fermi-Dirac statistics to the theory of para- and
ferromagnetic phenomena. His 1934 monograph Magnetism and Matter,
and his long series of research and review articles, became standard
references. His paper, 'The distribution of electrons among atomic levels',
written in 1924 at the age of 25, strongly influenced Pauli's formulation of
the exclusion principle a year later, and in part anticipated it. In 1 92 8 ,
stimulated b y some speculative ideas of Jeans o n 'liquid stars', Stoner's
interest turned briefly to astrophysics.
Jeans (1926, 1928) had suggested that material deep inside a star might
deviate from Boyle's law - a condition he deemed necessary for stabilityt -
t According to Jean's calculations, stars in radiative equilibrium would be vibrationally
unstable if they are gaseous throughout (a conclusion strongly qualified by later work).
He hoped thereby to characterize untenanted portions of the Hertzprung-Russell
diagram as regions of instability where jamming is relieved by the stripping-off of
successive layers of electrons. Not for some years was it recognized that stars whose
energy sources are concentrated toward the centre will have convective cores, for which
stability is not a problem (e.g. Jeffreys, 1930; Biermann, 1932; Cowling, 1934) .
due to jamming of incompletely ionized atoms of the heavier elements. This

idea at once ran into the difficulty that , for effective jamming, the size of the
inner electronic shells needed to be very much larger than radii estimated on
the basi s of Bohr theory and, later, quantum mechanics. Stoner ( 1929)
accordingly set out to examine whether significant deviations would result
from Fermi-Dirac degeneracy, which is a jamming of electrons in phase
space, and , specifically, what kind of upper bound this jamming imposes on
the density of stellar material . Stoner derived an upper bound (attained in
the limit of complete degeneracy) which varied as the square of the star's
mass and was about an order of magnitude larger than the mean density
typical of white dwarfs.
This work was quickly followed up by Wilhelm Anderson ( 1929) of Tartu
University, Estonia, who pointed out that near Stoner's upper limit the
electrons are approaching the speed of light, and the equation of state
therefore needs correction for the increase of particle mass with velocity.
Allowing for this, Anderson found that a completely degenerate star of given
density has a mass that is significantly lower than Stoner's non-relativistic
estimate. In fact , it is clear from his tabulation and formulae that his
equations predict the existence of a critical mass (close to one solar mass for
a star composed of helium) for which the density becomes infinite, and
above which no completely degenerate configurations exist. However,
Anderson refrains from any direct reference to this, and merely comments
that his equations must break down at high densities because they fail to
take account of effects such as the contribution of the gravitational potential
energy of the star to its total mass and the 'compressibility of the electrons
and protons' at very high densities.
Acknowledging Anderson's criticism, Stoner now attacked the problem
afresh, and laid the groundwork for a systematic treatment, in a paper,
submitted in December 1929, that contains what came to be known as the
Stoner-Anderson equation of state for a completely degenerate relativistic
gas. t Stoner ( 1930) explicitly confirms the existence of a limiting mass (his
estimate of 1 .7 solar masses is somewhat higher than Anderson's) 'above
t Nearly two years earlier the statistical thermodynamics of relativistic Bose and Fermi
gases had received a general and comprehensive formulation at the hands of Ferencz
Juttner ( 1928) of Breslau . However, Juttner did not explicitly consider the limit of
complete degeneracy (which greatly simplifies the general formulae) , and he was u naware
of the astrophysical relevance of relativistic degeneracy. He considered his work to be of
theoretical i nterest only. His paper appeared in a German physics journal and did not
come to the attention of astronomers for some years.
214 W. Israel
which the gravitational kinetic equilibrium considered will not occur'. On

the significance and implications of this he offers no comment beyond the
remark that the known white dwarfs fall below the critical limit.
In the calculations of Stoner and Anderson the astrophysical aspects were
handled very schematically : the sta r was simply idealized as a sphere of
uniform density. In view of the crudeness of this approximation it must be
considered fortuitous that their estimates for the critical mass came so close
to the correct value of 1 .44 solar masses (for a helium star). The definitive
value was obtained by the 19-year-old Subramanyan Chandrasekhar in July
1 930 (without knowing at the time of Stoner's work) during a sea voyage
from India to Cambridge. He showed his result to Fowler, who was
doubtful , and sent it to Milne for his opinion : the reaction was not
encouraging. It nevertheless did eventually find its way into print in the USA
as a brief paper in the Astrophysical Journal (Chandrasekhar, 193 la). As he
recalled in 1977 , 'I didn't understand at the time what this limit meant, and I
didn't know how it would end. But it is very curious that Fowler did not
think the result very important'.t
Although the concept of the atom as the indestructible building block of
matter had broken down , the belief remained firm that there must be a
'maximum density of which matter is capable', now generally supposed to
be 'given by known physical properties of material such as the radii of
electrons and nuclei' (Milne, 1930; Fowler, 1926).t Under the influence of
these ideas, Chandrasekhar ( 193 lb, 1932) for a while attempted to show that
white dwarfs can have arbitrarily large masses if they contain central cores in
which matter has reached nuclear densities and become incompressible.
This idea encounters numerous difficulties and it did not survive the
discovery of the neutron a year later.
In January 1932, Lev Landau submitted a short paper, entitled 'On the
theory of stars' , in which he independently recovered Chandrasekhar's
limiting mass. Landau's work has great sweep but there are some oddities.
He proposes that
. . . to attack the problem of stellar structure by methods of theoretical
physics . . . we must at first investigate the statistical equilibrium of a
given mass without generation of energy, . . . the quantum effects for the
t Chandrasekhar's reminiscences (Chandrasekhar, 1969, 1977 ; Wali, 1 982) offer a

fascinating behind-the-scenes glimpse into the unfolding of the dramatic developments of
the 1930s.
t The evolution of scientific views on an 'ultimate' density of matter has been traced by
W. Anderson ( 1936) and E. R. Harrison ( 1972).
Dark stars: the evolution of an idea 21:
non-relativistic Fermi-gas . . . lead to the existence of a stable

equilibrium . The presence of sources would produce only an additional
expansion due to the radiation pressure.
From this and his subsequent remarks it would appear that Landau had
entirely overlooked the key role of thermal gas pressure in supporting the
weight of hot stars. Apparently he was at this time under the impression that
the mass limit which he proceeded to derive for cold stars would apply to all
stars, at least to a good approximation .
Thus we get an equilibrium state only for masses greater [sicJ than a
critical mass M 0 = . . . about 1 . 5 0 (for m = 2 protonic masses). For
M > M 0 there exists in the whole quantum theory no cause preventing
the system from collapsing to a point . . . . As in reality such masses exist
quietly as stars and do not show any such ridiculous tendencies we must
conclude that all stars heavier than 1 .50 certainly possess regions in
which the laws of quantum mechanics (and therefore of quantum
statistics) are violated .
(Landau , 1932)
During the next three years, as Chandrasekhar completed the first of his
monumental contributions to astrophysics - a comprehensive theory of the
internal structure of white dwarf stars and of stars with degenerate cores
based on the Stoner-Anderson formula - the issue of the critical mass
continued to cast a pall over the whole subject. In October 1934 he
summarized the state of knowledge as follows.
Finally, it is necessary to emphasize one major result of the whole
investigation, namely, that it must be taken as well established that the
life-history of a star of small mass must be essentially different from . . . a
star of large mass. For a star of small mass the natural white-dwarf
stage is an initial step towards complete extinction . A star of large mass
( > .fi) cannot pass into the white-dwarf stage and one is left speculating
on other possibilities.
(Chandrasekhar, S., 1 934b)
One possibility in particular was engrossing his thoughts at the time .
Coronal evaporation from Wolf-Rayet and similar stars, first noted by the
Canadian astronomer, Carlyle S. Beals ( 1929), at the Dominion
Astrophysical Observatory, had become an active focus of interest, and
Chandrasekhar ( 1934a) himself had worked on the theory. Earlier that year
he had presented a report to the regular jo u rnal seminar conducted by
Eddington on a paper by Kosirev ( 1934) dealing with physical conditions in
the extended photospheres of Wolf-Rayet stars. At the end of his lecture he
21 6 W. Israel
ventured - in violation of the rules : speakers were not supposed to discuss

their own work ! - to slip in the suggestion that mass shedding might be the
safety valve that provides relief for overweight stars. Eddington, however,
could not accept this as a panacea. In conversations with Chandrasekhar, he
repeatedly expressed the conviction that, at least in some cases, evolution
must continue to the point where pressure support failed and the star was
obliged to contract .
Chandrasekhar outlined his tentative hypothesis in the concluding
section of a paper submitted to the Monthly Notices on 1 January 193 5 :
A possibility i s that at some stage in the process of contraction the flux
of radiation in the outer layers of the stars will become so large that a
profuse ejection of matter will begin to take place. This ejection of
matter will continue till the mass of the star becomes small enough for
central degeneracy to be possible . . . . Thus, whether the star is of large
or small mass, the final stage in its evolution is always the white-dwarf
stage . . . .
(Chandrasekhar, 1935a)
This scenario, at first advanced and treated with reserve (e.g . Russell ,
Dugan and Stewart, 1938), gradually took root over the next two decades
(e.g . Gamow· and Schonberg, 194 1 ; Chandrasekhar, 195 1 ; Hoyle, 1955 ;
Payne-Gaposchkin , 1957 ; Schatzman , 1958; Struve, 1962 ; O'Dell, 1963) . It
finally developed into a widespread implicit faith (especially outside the
Soviet Union) in continuous or catastrophic mass ejection as a universal
regulating mechanism which brings every star in its final stages below the
critical weight. From 1963, when it became attractive to contemplate other
possibilities, the notion was viewed more critically :
The conventional view is that in all such cases a process of mass ejection
takes place, conveniently reducing the final mass below one or other of
the mass limits just mentioned. We wish to emphasize that this view is
no more than a superstition . . . . Indeed, it would be a curious situation
if astrophysical processes occurring long before the onset of the final
implosion crisis were to operate always to prevent the crisis from
arising. This would imply an unlikely 'foreknowledge' on the part of
natural processes.
(Hoyle, Fowler, Burbidge and Burbidge , 1964)
But we are running ahead of our story.
7.5 Eddington's intervention (1935)

There is evidently need for a re-survey of this subject, which had
become confused by the distractions of the Stoner-Anderson formula.

(Eddington, 1939)
Early in 1935, as Chandrasekhar was quietly putting the finishing touches to
his work on white dwarfs, there was a bombshell. Eddington , whose time
had been occupied with the development of his Fundamental Theory, quite
unexpectedly entered the arena with a characteristically dramatic opening
salvo.
Dr Chandrasekhar has been referring to degeneracy. There are two
expressions used in this connection , 'ordinary' degeneracy and
'relativistic' degeneracy, and perhaps I had better begin by explaining
the difference. They refer to formulae expressing the electron pressure P
in terms of the electron density a . For ordinary degeneracy P = K a513 •
But it is generally supposed that this is only the limiting form of a more
complicated relativistic formula, which shows P varying as something
between a513 and a413 , approximating to a413 at the highest densities. I
do not know whether I shall escape from this meeting alive, but the
point of my paper is that there is no such thing as relativistic
degeneracy !
I would remark first that the relativistic formula has defeated the
original intention of Prof. R. H . Fowler . . . . In 1924 . . . I was troubled
by a diffi�ulty that there seemed to be no way in which a dense star
could cool down . . . . Soon afterwards Fermi-Dirac statistics were
discovered, and Prof. Fowler applied them to the problem and showed
that they solved the difficulty ; but now Dr Chandrasekhar has revived
it again . Fowler used the ordinary formula ; Chandrasekhar, using the
relativistic formula which has been accepted for the last five years,
shows that a star of mass greater than a certain limit .A remains a
perfect gas and can never cool down . The star has to go on radiating
and radiating and contracting and contracting until, I suppose, it gets
down to a few km radius, when gravity becomes strong enough to hold
in the radiation, and the star can at last find peace.
Dr Chandrasekhar has got this result before, but he has rubbed it in
in his last paper; and , when discussing it with him, I felt driven to the
conclusion that this was almost a reductio ad absurdum of the relativistic
degeneracy formula. Various accidents may intervene to save the star,
but I want more protection than that. I think there should be a law of
Nature to prevent a star from behaving in this absurd way !
. . . I feel satisfied myself that the current formula is based on a partial
relativity theory, and that if the theory is made complete the relativity
218 W Israel
corrections are compensated, so that we come back to the 'ordinary'

formula.
Suppose we are dealing with a cubic centimetre of material in the
middle of a star. Ordinarily we analyse this into electrons, protons, etc.,
travelling about in all directions . . . . In the ordinary analysis of matter
into electrons one is dealing with progressive waves; but in the analysis
which leads to the Exclusion Principle (used in deriving the degeneracy
formula) the electron is represented by a standing wave . . . . The
electron represented by a progressive wave can be brought to rest by a
Lorentz transformation, and it then becomes a standing wave. This
transformation introduces a factor into the equation , which is not
needed if the waves referred to are standing waves originally . . . .
(Eddington, 1935a)
It is not difficult to recapture the spell cast by Eddington's charisma on the
audience in Burlington House on 1 1 January (Wali, 1982). But in cold print
his logic must have appeared somewhat less pellucidly clear.
Eddington never wavered in his rejection of the Stoner-Anderson
equation of state, though his ground for opposing it seemed to shift this way
and that in the half-dozen or so publications he devoted to this subject over
the next few years (e.g. Eddington, 1935b, 1939, 1940). I shall not attempt to
recapitulate his various arguments here; Schatzman ( 1958) has summarized
and critically analysed some of them.
Apart from some efforts at rebuttal (e.g. M0ller and Chandrasekhar,
193 5 ; Peierls, 1936; Dirac et al. , 1942), Eddington's objections attracted
little notice in the literature. The comprehensive review articles of Hund
(1936) and Stromgren ( 1937), the standard astronomy text of the day
(Russell, Dugan and Stewart, 1 938) and Gamow's ( 1 937, 1 940) books refer
to the Stoner-Anderson formula and the Chandrasekhar limit without
mentioning Eddington's reservations. In the 1936 edition of his massive
book, Statistical Mechanics, R. H . Fowler, whose non-relativistic formula
Eddington was stoutly and gratuitously 'defending', presents the standard
derivation with a courteous nod to Eddington in a footnote :
If his contentions are correct then the formulae derived from [the
relativistic relation between energy and momentum] are meaningless
and we must always use formulae derived from E = p;/2m (i.e.
Schrodinger's equation) for electrons represented by standing waves in
an enclosure.
The events of 1935 have special significance for the light they shed on the
generally confused state of the subject at the time and because of
Eddington 's key role and position of influence in the development of

twentieth-c entury astrophysic s - 'our incomparable pioneer' in the words of
Milne's obituary. With his unrivalled command of the two key theories ,
stellar structure and general relativity, it would seem that Eddington was
uniquely placed at this critical juncture to point the way to future
developments - to predict the existence of a new class of celestial objects. But
events took a different turn . The roots and repercussions of this curious and
sad episode will inevitably remain a focus of speculation for years (Sullivan,
1979 ; Wali , 1982 ; Blandford, this volume, Chapter 8) . The questions
involved are complex and subtle and may never be fully resolved. Here I
shall venture only some scattered comments that appear relevant to the
issue.
The scenario which seemed a reductio ad absurdum to Eddington was not
yet the ultimate catastrophe of collapse to zero volume that confronts
physicists today. As we shall see, Eddington, like virtually every relativist of
the time, considered the Schwarzschild radius to be both a singularity and
an impassable barrier. The image that he conjures up of the star 'at last
finding peace' is of a body frozen at the Schwarzschild radius ; it was not yet
appreciated that a co-moving observer would see something radically
different.
No one was better placed than Eddington to recognize that asking for an
equation of state that would in principle prevent the collapse of an arbitrary
configuration is asking the impossible. It had been known since the time of
Schwarzschild ( 19 1 6b) that, once general relativity is taken into account , a
sphere of radius smaller than li" times its Schwarzschild radius cannot hold
itself in equilibrium even if it is made of incompressible material . (Only a few
months earlier Eddington had communicated to the RAS a paper (Sen ,
1934) on this very subject which made use of Eddington's own definition of
'incompressibility' (see below).) However, if, as Eddington believed,
Fowler's equation of state remains accurate at high densities, white dwarf
radii would all be much larger than their Schwarzschild radii unless they
happen to be more than 100 times heavier than the sun, and in the 1930s no
stars as massive as this were definitely known to exist . Thus, Eddington
believed that no actual star would ever get into a situation where general
relativistic effects could become significant.
There is evidence suggesting that Eddington was set on a collision course
with the Stoner-Anderson formula even if he had not been put off by its
consequences. This actually goes back to an unorthodox definition of
particle density which he proposed in his 1923 book, The Mathematical
220 W Israel
Theory of Relativity (p. 12 1). The presently accepted formulation , first

clearly set forth by van Dantzig (1939), considers n to be the magnitude (or
temporal component in the rest-frame) of a conserved 4-vector Nµ, the
particle flux vector. Eddington, however, believed that n should be
extractable from the stress-energy tensor Tµ v (at least for a monatomic
system) and proposed the identification nm = T�, where m is the particle rest
mass. This idea, enshrined as it was in a major textbook, remained a source
of confusion for years (e.g. Sen , 1934; Curtis, 1950). From a conventional
standpoint it is plainly wrong, since it would require the internal energy
density T8 - nm to be three times the pressure, which is too large by a factor
of two in the case of a non-relativistic gas. In his later work Eddington
( 1936a) was well aware of this discrepancy , but believed it could be
attributed to an extra potential or 'interchange' energy associated with an
effective repulsive 'force' describing the action of the Pauli exclusion
principle. (Weisskopf ( 1975) has recently described how the concept of such
a phenomenological exclusion 'force' offers a powerful heuristic viewpoint
for a semi-quantitative understanding of the bulk properties of matter.)
To bring out the connection with Eddington's stand on the Stoner
Anderson formula, let us evaluate Tµ v and N v for a gas. A particle with 4-
velocity vµ has the ordinary velocity dxµ/dx 0 = y - 1 vµ with respect to an
arbitrarily chosen Lorentz frame, where y = v 0 = ( 1 - v 2 ) - 1 12 is the Lorentz
dilation factor. (We have set c = l .) According to the conventionally
accepted formulation, Tµ v and Nµ are defined as the fluxes of 4-momentum
mv v and particle number, i .e.
Tµ v = L y - 1 vµmvv , Nµ = L y - 1 vµ.
The summation is over all particles in a cell of unit volume which is a t rest in
the chosen frame. {The factors y - 1 , which appear to spoil Lorentz
covariance, actually guarantee it : y - 1 d 3p is the Lorentz-invariant measure
of volume in momentum space.) Specializing to the rest-frame of the
medium we correctly obtain n = N° = L 1 .
Eddington , by contrast, always held that Tµ v is to be calculated by
summing, in the macroscopic rest-frame, the stress-energy tensors of the
individual particles. This prescription yields a different tensor, T�<ld =
L mvµvv , whose trace, (T�)Edd = m L 1 = nm, indeed determines the particle
density, as Eddington claimed .
The effect of Eddington's proposal is to exaggerate the stability of massive
stars : absence of the contraction factor y - 1 means that , for a given particle
density, his stress tensor predicts a higher pressure than the conventional
one. Specifically, consider an electron gas in its lowest quantum state : each
of the lowest quantum cells (with phase volume h 3 ), up to a momentum
ceiling p * (say), will be filled by a pair of electrons having opposite spins. The
summation I is thus to be replaced by the integral s�· . . . 2(4np 2 dp )/h 3 •
Eddington's expressions then give P Edd ,....., p ; , n ,....., P! , i .e. P Edd ,....., n 51 3 , which is
precisely Fowler' s formula.
Eddington's own presentation of his arguments never took quite this
simple form, and he never referred to their 1923 pre-quantum antecedents,
presumably because he considered that enumeration of quantum states is a
procedure different in principle from the counting of classical particles. t
Nevertheless, his classical and quantum procedures conform to the same
mould. (The resemblance is plainest in the 1 940 paper.)
I have gone into this point in painstaking detail because it seems to have
gone generally unnoticed and because it reveals the underlying continuity
and inner consistency of Eddington's thought over a time-span stretching
well before the events of 1935. By his own account (Eddington , 1936b), the
motivation which first caused him to question the Stoner-Anderson result
was the 'stellar buffoonery' to which it led ; but there can be little doubt that
the reason for his sustained opposition was grounded in purely technical
considerations whose seeds went back more than a decade. Indeed, it is
interesting that after its debut as the launch-pad of his 193 5 paper, the
contraction scenario and its 'absurdity' make no further appearance in his
published work .
To speculate about a hypothetical replay of the famous RAS talk , with all
reference to a 'reductio ad absurdum ' magically removed , is idle but hard to
resist (Chandrasekhar, 1969, 1972 ; Sullivan, 1979 ; Wali, 1982). Could it
have changed astrophysical history? Given the limited optical window of
pre-war astronomy and the many psychological hurdles, the possibility
appears remote of any radical shift in pre-war opinion on this issue.
In the first instance, there were Eddington's own preconceptions. By 1936
his unconventional stress tensor had so permeated his thinking on 'molar
relativity' and Fundamental Theory that a change of course would have
meant dismantling a complex interlocking structure. Attempts to rebut his
contentions must have been a source of deep frustration to both sides, since
they were often speaking different languages. In the manner typical of
creative minds, his way to assimilate a new body of ideas, such as relativity
or quantum mechanics, was to re-create it in his own terms. It made
tSee the remarks on p. 254 of Eddington ( 1936a).

222 W Israel
communication not always easy. Once, defending the Eddington model for
main-sequence stars against a rival theory of Milne's, he declared,
Prof. Milne did not enter into detail as to why he arrives at results so
widely different from my own ; and my interest in the rest of the paper is
dimmed because it would be absurd to pretend that I think there is the
remotest chance of his being right . . . . I maintain that the interior
determines the state of the photosphere ; Prof. Milne somehow reaches
the opposite view.
(Eddington, 1929)
It stung Milne to the retort :
I recognize that Sir Arthur Eddington has dug a most valuable trench
into unknown territory. But he has encountered a rocky obstacle which
he cannot get round. If he would make the mental effort to scramble up
the sides of the trench he would find the surrounding country totally
different from what he had imagined and the obstacle enti rely an
underground one.
(Milne, 1930)
Suppose, nevertheless, that Eddington's technical concerns could have
been sufficiently allay�d to make the Stoner-Anderson formula and the
Chandrasekhar limit logically acceptable to him. Would his belief in the
contraction scenario as an inescapable consequence have held firm? This is
the hardest and most hypothetical question of all : we do not know how deep
his convictions on this issue really went. Others who accepted the
Chandrasekhar limit did not feel impelled to this extremity. He himself
speaks of 'various accidents intervening to save the star'. One has to judge
whether it is likely that, lacking any observational support for collapse, he
would have been prepared to abandon an astronomically conservative
stance to take a position that even the arch-radical Landau at that time
considered untenable. If the inner conviction was there, it is more than
probable that he would have done so. The courage and integrity with which
he defended his beliefs were legendary. Eddington's failure in 1935 was not a
failure of nerve, but an aberration of a soaring imagination.
One is on much safer ground in predicting the probable reaction of his
contemporaries. In 1935, the astronomical community was not yet ready to
'buy' the idea of gravitational collapse, not even if a master salesman like
Eddington had been ready to exert all of his persuasion. This is plain from
the apathy that greeted Oppenheimer's proposal of the idea four years later.
Even today, after half a century, it would be rash indeed to claim that the
black hole's battle for acceptance is finally over.
The year 1935 marks the end of a glorious period in British astronomical
history. For an unbroken span of 20 years the meeting rooms of the Royal
Astronomical Society in Bu rlington House had held the centre of the world's
astrophysical stage. But now the spotlight was moving westward .
7.6 Neutron stars and gravitational collapse (1934-59)

Once man is in a rut he seems to have the urge to dig ever deeper.
(Zwicky, 1957)
It was natural enough that Chadwick's discovery of the neutron in February
1932 should invite speculation about stars consisting wholly or partly of
neutrons. The question was indeed quickly taken up by a few adventurous
spirits among the astronomers and the nuclear physicists, though with
entirely different motives. Conjectures about the nature of such objects
accordingly took two quite distinct forms during the 1930s.
The astronomically motivated view was that neutron stars are remnants
of supernovae (and hence very rare objects). The originator of this view was
the volatile Caltech astronomer Fritz Zwicky (born in Bulgaria and
educated in Switzerland), who remained for 25 years its sole advocate in the
face of a conventional wisdom which maintained that astronomy does not
need anything more exotic than white dwarfs as endproducts of stellar
evolution.
The second line of speculation was prompted by the obstacles faced by
nuclear physicists in the mid- 1930s in their attempt to account for stellar
energy generation in terms of thermonuclear reactions. The idea, promoted
mainly by Landau ( 1 938) and Gamow ( 1937), was that many or possibly all
stars contain degenerate neutron cores. It was supposed that a slow growth
of the core fuels the star's radiation , as normal material becomes
neutronized and drastically reduced in volume, thus releasing gravitational
energy . {That there were reasons to expect that neutronization would
actually be a runaway process, attended by a rapid collapse of the central
regions of the star, was not appreciated until after the work of Gamow and
Schonberg (194 1 ), who argued that cooling due to neutrino losses by the
'urea-process' in inverse /3-decay would result in a catastrophic failure of
pressure support near the centre.) With Bethe's discovery of the carbon cycle
in 1939 , this line of thinking came to an abrupt end . But it did survive just
long enough to spawn the classic contributions of Robert Oppenheimer and
his students George Volkoff and Hartland Snyder at the Berkeley campus of
the University of California.
It was, however, the astronomers who introduced both the concept and
224 W. Israel
the term 'neutron star' in its presently understood sense. It came in a brief
paragraph, appended as an 'additional remark', to a visionary pair of papers
by Baade and Zwicky in which supernovae were named and distinguished as
a class of objects fundamentally different from ordinary novae :
In addition, the new problem of developing a more detailed picture of
the happenings in a supernova now confronts us. With all reserve we
advance the view that a supernova represents the transition of an
ordinary star into a neutron star, consisting mainly of neutrons. Such a
star may possess a very small radius and an extremely high density. As
neutrons can be packed much more closely than ordinary nuclei and
electrons, the 'gravitational packing' energy in a cold neutron star may
become very large, and under certain circumstances, may far exceed the
ordinary nuclear packing fractions. A neutron star would therefore
represent the most stable configuration of matter as such. The
consequences of this hypothesis will be developed in another place,
where also will be mentioned some observations that tend to support
the idea of stellar bodies made up mainly of neutrons.
(Baade and Zwicky, 1934)
Zwicky ( 1938 , 1939) was well aware that general relativity was required
for a detailed understanding of neutron star structure. But his attempts to
develop a general relativistic theory were overtaken by the definitive (and
differently motivated) work of Oppenheimer and Volkoff ( 1939).
It should be mentioned here that the general idea that stellar explosions
were associated with collapse to a superdense configuration was not original
with Baade and Zwicky but had been in the air for several years (e.g. Menzel ,
1926 ; Milne, 193 1) , and appeared to derive observational support from the
resemblance of old novae and the nuclei of planetary nebulae to white d warf
stars (e.g. Gerasomovic, 193 1). Perhaps the earliest reference to the idea is to
be found in a paper by H. N. Russell ( 1925). This was a last-ditch attempt to
reconcile the giant-and-dwarf evolutionary theory with Eddington's mass
luminosity relation, but it anticipates in a very generalized yet remarkably
judicious way, considering the state of nuclear physics at the time, a number
of modern ideas about the ignition of thermonuclear reactions in stars and
about stellar evolution .
It would appear that Lev Landau began to think about neutron stars on
the very day that news of Chadwick's discovery reached Copenhagen . Leon
Rosenfeld ( 1973) has recalled how, on that evening, as he, Bohr and Landau
sat discussing its various implications, Landau came out with the idea of
'weird stars' (unheimliche Sterne). But it was five years before Landau
committed any of his thoughts to paper. In a brief article entitled 'The origin
of stellar energy' he pointed out that
. . . in spite of the fact that the 'neutronic' state of matter is, in usual
conditions, energetically less favourable, since the reaction of neutron
formation is strongly endothermic, this state must nevertheless become
stable when the mass of the body is large enough. In this case the
gravitational energy gained in going over to the neutronic state with its
greater density, compensates the losses of internal energy. t
Landau 's estimate of 10- 3 M 0 for the minimum mass of a neutron core
(using Fowler's non-relativistic equation of state) was later revised to a few
tenths of a solar mass by Oppenheimer and Serber ( 1938). His motivation
emerges in the concluding paragraphs :
. . . we can regard a star as a body which has a neutronic core whose
steady growth liberates the energy which maintains the star at its high
temperatures . . . .
. . . the author has shown in a previous article [Landau , 1932] that
the formation of a core mu st certainly take place in a body with a mass
greater than 1 . 5 0 . In stars with smaller mass the conditions which
could make possible the formation of the initial core have yet to be
made clear.
(Landau , 1938)
Similar sentiments were expressed by Gamow ( 1937) in the final chapter
of his book, Atomic Nuclei and Nuclear Transformations, although at the
time of writing the outlook for a thermonuclear origin of stellar energy was
not as dim as it came to seem two years later. The situation as it appeared in
the fall of 1938 was summarized by Oppenheimer and Robert Serber :
It would seem that the formation of deuterons by proton collision , and
at the least partially regenerative capture of protons by elements
between carbon and oxygen could be made to account successfully for
the main-sequence stars . . . . Nevertheless it has been clear that these
reactions could in no way account for the enormously greater radiation
of such stars as Capella, and for these one would either have to invoke
other and readier nuclear reactions with a correspondingly reduced
time-scale, or one would be led, as in the earlier arguments of Milne, to
expect serious deviations from the Eddington model.
t T. E. Sterne ( 1933) had shown earlier that, even without taking gravitational energy
into account , transition to the neutronic phase is energetically favourable for cold matter
compressed to densities exceeding 101 g/cm 3 .
0
226 W Israel
In this connection . . . the suggestion [Gamow, 1937 ; Landau , 1938]

of a condensed neutron core, which would make essential deviations
from the Eddington model possible even for stars so light that without a
core a highly degenerate central zone could not be stable, still seems of
some interest.
It was against this background that Oppenheimer and his students
embarked upon their remarkable investigations. The first question of
interest was 'to investigate whether there is an upper limit to the possible size
of such a neutron core' . It was obvious that Newtonian gravitational theory
would give a generous upper limit of four times the Chan � rasekhar limit for
white dwarfs, or about six solar masses, since the star's pressure would be
carried by neutrons instead of mainly by the electrons stripped off from
helium nuclei, thus reducing the mean mass per pressure - producing
particle by a factor 2. However, because of the much smaller radius of a
neutron star, general relativistic corrections would become appreciable , as
would the weight associated with the kinetic energy of the particles, and
both effects could be expected to bring down this upper limit substantially.
In their paper, 'On massive neutron cores', Oppenheimer and V olkoff ( 1939)
reported the results of their numerical integrations of the general-relativistic
equations of hydrostatic equilibrium for various central pressures. For
masses larger than 0.7 solar masses - the 'Oppenheimer- Volkoff limit' -
there were no equilibrium configurations.
Since almost nothing was known at the time about nuclear forces,
Oppenheimer and Volkoff had little choice but to use the Stoner-Anderson
equation of state for a gas of free neutrons in their calculations. But they
went on to show that their estimate for the limiting mass would not be
appreciably affected by any modification of this equation at supranuclear
densities that was physically reasonable in the sense that the trace of the
stress-energy tensor did not become negative. t
Six months later, on 10 July 1939, Oppenheimer and Snyder followed this
with the submission of a manuscript that has strong claims to be considered
the most daring and uncannily prophetic paper ever published in the field .
Its scope and breathtaking sweep may be gathered from the abstract :
When all the thermonuclear sources of energy are exhausted a
sufficiently heavy star will collapse. Unless fission due to rotation, the
radiation of mass, or the blowing off of mass by radiation, reduce the
star's mass to the ord�r of that of the sun , this contraction will continue
t This was widely considered to be a rigorous requirement, whose proof was attributed
to von Neumann (cf. Chandrasekhar, 1935b). However, Zel'dovich ( 196 1) has produced
a counter-example showing that it is possible, at least in principle, to circumvent it.
indefinitely . . . the radius of the star approaches asymptotically its

gravitational radius ; light from the star is progressively reddened , and
can escape over a progressively narrower range of angles . . . . The total
time of collapse for an observer co-moving with the stellar matter is
finite, and for . . . typical stellar mass of the order of a day ; an external
observer sees the star asymptotically shrinking to its gravitational
radius.
There is nothing in this paper which needs revision today, even its
terminology is undated . Nevertheless, the questions it left unanswered were
to retard acceptance of the results when interest in the subject began to
revive 20 years later.
Oppenheimer and Snyder had integrated the interior spherically
symmetric Einstein equations for a case in which they could obtain an
analytic solution, in the (as later work was to confirm, correct) belief that
'the general features of the solution obtained in this way give a valid
indication even for the case where the pressure is not zero, provided that the
mass is great enough to cause collapse'. They showed how this interior
solution, expressed in co-moving coordinates, could be joined to the exterior
Schwarzschild solution, expressed in the standard Schwarzschild
coordinates. Unfortunately, with this choice of coordinates, it was
impossible to obtai'n a global picture of the dynamics beyond the point
where the star reaches its gravitational radius, since here the exterior time
becomes infinite. To a reader of the paper, it was thus left obscure how the
interior co-moving picture, in which the collapse proceeds to zero radius in
a finite proper time, is to be reconciled with the external view , which sees the
contraction 'freezing' at the gravitational radius. t The authors do not even
make clear their views on the status of the gravitational radius, which was
still generally considered to be some kind of singularity. The nature of the
Schwarzschild 'singularity' was being discussed by H . P . Robertson and
Einstein at Princeton early in 1939. Robertson's unpublished investigations
(to be described in the following section) were carried out in 1936-1937
while on a sabbatical leave at Caltech. It is thus conceivable, though I have
no evidence for it, that Oppenheimer had learnt from Robertsont about the
t A clear global picture first emerged from the work at Princeton of Misner's student
Beckedorff ( 1 962), who fitted the interior and exterior geometries together into a single
spacetime diagram.
t Opportunities for personal contact between the two men were certainly available.
Robertson's host at Caltech was Richard C. Tolman, a friend and colleague of
Oppenheimer. Tolman was then engaged in completing his book, The Principles of
Statistical Mechanics ( 1938), dedicated to Oppenheimer, who was consulted on its every
aspect .
228 W. Israel
fictitious character of the singularity, and may even have supposed that this
was common knowledge in relativistic circles.
By the time the Oppenheimer-Snyder work appeared in print in
September 1939, Great Britain and France were at war with Germany. The
authors had exhausted the topic as far as it could be pursued in the prewar
'optical' era of astronomy. No tie-up with observation was apparent in the
scenario of a star that is invisible to begin with, and at the end of its nuclear
I
resources, undergoing a total collapse. Even the premises on which the

research was based had been undermined by the appearance, six months
earlier, of Bethe's ( 1939) work, 'On energy genera ti on in stars', which had
dealt the finishing blow to ideas that stars need neutron cores i n order to
shine.
In the months that remained before they dispersed, the three authors
moved on to more fashionable problems in nuclear and cosmic ray physics
and meson theory. Early in 1940 Volkoff left for Princeton to work with
Wigner on the theory of nuclear forces, then returned to the University of
British Columbia to take up a teachin·g position ; Snyder accepted a position
at Northwestern University (Illinois); by the end of 194 1 Oppenheimer had
been recruited into the Manhattan Project. They never returned to active
research in this field, though Oppenheimer sometimes attended the 'Texas'
symposia on relativistic astrophysics in later years.
By 1940 the subject had entered its dark age. Interest remained active in
the possible ignition mechanisms of supernovae (e.g. Gamow and
Schonberg , 194 1 ; Schonberg and Chandrasekhar, 1 942; Hoyle, 1 946), but
the dynamics of the explosion was beyond the computational facilities then
available. It was generally supposed that the remnant would be a white
dwarf (e.g. Hoyle, 1955). Minkowski's ( 1 942) analysis of the data for the
Crab nebula, which pointed to a radius for the central star about twice that
of the earth, could be seen as lending support to this view. Only the maverick
Zwicky (e.g. 194 1 , 1957) still professed an interest in neutron stars, and he
continued to press his views from time to time, but nobody paid attention. t
Post-war strides in computer technology and the understanding of
nuclear forces eventually made it timely to reconsider the Oppenheimer
Volkoff problem using a more realistic equation of state. In 1957 this was
undertaken by the Princeton nuclear physicist John Archibald Wheeler and
t In his popular book of 1940, Gamow devoted considerable space to Zwicky's ideas,
but became noncommittal on this issue in his subsequent technical papers (Gamow and
Schonberg, 194 1 ; Gamow, 1944).
his assistants, B. Kent Harrison and Masami Wakano. It marked the entry
of a charismatic new personality into the then moribund field of general
relativity . Wheeler's interest was not primarily in the possibility of deriving
observable predictions. He was disturbed by a question of principle - 'What
is the final equilibrium state of an A-nucleon system when A is large?' - and
he found the Draconian answer offered by Oppenheimer and Snyder difficult
to accept.
Assembling all the available theoretical information, Harrison and
Wheeler constructed a semi-empirical equation of state for 'cold matter,
catalysed to the endpoint of thermonuclear evolution' at all stages of
compression from 10 g/cm 3 to supranuclear densities. Using this, Wakano
performed 44 numerical integrations, for different choices of central density,
of the general-relativistic stellar equilibrium equations on Princeton's
MANIAC computer. The resulting plot of mass vs. central density was a
curve with two humps, corresponding to the white dwarf and neutron star
configurations. The Chandrasekhar and Oppenheimer-V olkoff results were
confirmedt and for the first time brought together into a single overall
picture.
Wheeler's report on this work to the Solvay congress in Brussels on J ..
June 1958 was only part of a panoramic survey of the entire field that did not
shrink from dragging basic issues into the open :
Of all the implications of general relativity for the structure and
evolution of the universe, this question of the fate of great masses of
matter is one of the most challenging. Moreover, the issue cannot be
escaped by appealing to stellar explosion or rotational disruption, for
the issue as it presents itself today is one of principle, not one of
observational astrophysics.
Won't the star explode? Let it! . . . Simply catch the ejected matter
and extract its kinetic energy . Let it fall back on the star. Then the
original number of nucleons, A, is restored ; but the mass-energy of the
system drops. Ultimately the star gets tired. It can't eject matter. It
can't radiate photons. It can't emit neutrinos. It comes into the
absolutely lowest state possible for an A-nucleon system under the dual
action of nuclear and gravitational forces . . . this is the state we are
interested in as a matter of principle.
t Equations of state currently considered viable give mass limits three or four times
higher than the o riginal Oppenheimer-Volkoff value (Baym and Pethick, 1979).
230 W Israel
It may not be necessary to found this discussion entirely on idealized

experimentation and issues solely of principle. It is conceivable that
astrophysical evolution may occasionally or even often lead to
contracting stars substantially more massive than the sun.
Regarding the 'final equilibrium state' he remarked :
Perhaps there is no final equilibrium state : this is the proposal of
Oppenheimer and Snyder . . . .
A new look at this proposal today suggests that it does not give an
acceptable answer to the fate of a system of A-nucleons under
gravitational forces . . . .
No escape is apparent except to assume that the nucleons at the
centre of a highly compressed mass must necessarily dissolve away into
radiation - electromagnetic, gravitational , or neutrinos, or some
combination of the three - at such a rate or in such numbers as to keep
the total number of nucleons from exceeding a certain critical number.
Oppenheimer, who was present at the meeting, demurred :
Would not the simplest assumption about the fate of a star more than
the critical mass be this, that it undergoes continued gravitational
contraction and cuts itself off from the rest of the universe?
But Wheeler then, and for several years afterward , had strong
reservations. He pointed out that in the Oppenheimer-Snyder scenario for
the spherical collapse of dust in vacuo
( 1) No mechanism of release of the gravitational energy into the
surroundings is taken into account . . . . Therefore this approach rules
out by definition any approach to an equilibrium.
(2) The particles are envisaged as falling into a K. Schwarzschild
singularity. However . . . the forces between nucleons come in in a most
vital way . any answer is incomplete that does not deal with the ultimate
. .
constitution of a nucleon.
(Adams et al., 1958)
Five or six years later, when it became commonplace to consider seriously
the collapse of astrophysical objects 108 times as massive as the sun, it was
easy to allay these concerns. For masses of this order, as Zel'dovich and
Novikov ( 1965) emphasized , densities are only a few g/cm 3 at the time of
crossing the gravitational radius :
Under these conditions, which are in no way remarkable, certainly
nothing fantastic can take place. The only thing that is unusually large
is the gnrvitational field, but according to the principle of equivalence
the gravitational field itself does not produce local changes in the laws
that govern physical processes.
By the early 1960s the Soviet school had established a commanding

position in this rapidly moving field. This development owes much to the
towering figure of Ya. B . Zel'dovich. But an important contributory factor
was the headstart in psychological preparedness which the Soviet scientists
held on their Western colleagues. In Western circles the work of
Oppenheimer and Snyder was a forgotten skeleton in the cupboard by the
1950s, and the notion of gravitational collapse, if anyone had even thought
of raising it , would have been dismissed as the wildest speculation ; but in the
Soviet Union it was an orthodox textbook result. In the 195 1 edition of their
widely read and admired treatise, Statistical Physics - the English
translation did not become available until after 1958 - Landau and Lifshitz,
citing Oppenheimer and Snyder, stated the position clearly and
unequivocally :
The question arises of the behaviour of a body with mass greater than
Mrnax · It is clear from the start that such a body must tend to contract
indefinitely . . . . From the point of view of a distant observer . . . the
sphere contracts so that its radius tends asymptotically to . . . its
gravitational radius. From the point of view of a 'local' observer the
substance 'collapses' with a velocity approaching that of light , and it
reaches the centre in a finite proper time.
Within a year of the Solvay conference there were signs that Wheeler's
crusade was beginning to find echoes. The Canadian nuclear astrophysicist
A. G . W . Cameron announced that, in disagreement with the establishment
view , he had concluded from 'an examination of the physics of supernova
explosions and of the formation of the elements . . . that neutron stars are
probable products of the supernova process'. His paper concludes with the
words :
It may also be noted that we are still faced with the basic philosophical
dilemma associated with the finite number of nucleons contained in the
limiting models for which the general relativistic equations of
hydrostatic equilibrium admit a solution. As Wheeler ( 1958) has noted
for similar cases, despite the fact that there are no solutions for larger
numbers of nucleons, what is to prevent such nucleons from being
added to a neutron star?
(Cameron, 1959)
7.7 The Schwarzschild 'singularity' (1916--66)

There is something queer about a gravitating particle, it seems worthy
of further study.
(Synge, 1949)
232 W. Israel
With the lid prised open, numerous investigators, encouraged to a

certain extent by the work of Oppenheimer and Snyder on
gravitational collapse, rushed to explore the bewildering region r < 2m.
At the present time, the nature of bodies which have collapsed into the
interior region and their attendant phenomena appear familiar and
substantive to these investigators.
(Cooperstock and Junevicus, 1973)
Karl Schwarzschild ( 19 16a) obtained his famous metric, representing the
spherisymmetric gravitational field of a point mass, in December 19 1 5 , just
before Einstein's theory reached its definitive, generally covariant form. For
lack of incentive, the nature of the complete vacuum geometry described by
this metric was still not generally understood 40 years later, although there
had been scattered attempts to probe the 'bewildering region' r < 2m. The
final breakthrough in understanding, which came in the late 1950s, emerged
virtually by osmosis and was sparked less by concerns about gravitational
collapse than by Wheeler's ( 1955) bo�dly imaginative ideas on multiply
connected topologies in general relativity, based on his conviction that
'general relativity, quantized , leaves no escape from topological
complexities'. The scientists chiefly responsible were, like Wheeler,
specialists in ·other areas who were newcomers to the field of general
relativity, and thus unburdened with the traditional belief that r = 2m was
somehow 'unphysical' , a belief encoded in the very name 'Schwarzschild
singularity't and supported by two generations of textbooks and the
authority of Einstein himself. Indeed, many professional relativists and
theoretical physicists at first reacted conservatively to the new developments
(e.g . Synge, 1960; McCrea, 1964; Morrison, 1965, 1977), an attitude that
lingers in some quarters today.
t The Schwarzschild metric in its standard form
ds 2 =f - 1 dr2 + r 2 (d0 2 + sin 2 O d</> 2 ) -f dt 2 , f= 1 - 2m/r
exhibits singularities at r = O and at r = 2m (called the 'gravitational radius' for a particle
of mass c 2m/G). It is generally understood nowadays that, unlike r = 0 where the
curvature actually becomes infinite, the 'Schwarzschild singularity' at r = 2m is merely an
apparent effect (like the singularity of polar coordinates at the origin) due to a break
down of Schwarzschild's time coordinate at the gravitational rad ius. One symptom of
this breakdown is that, although light rays moving radially inwards or outwards cross
the gravitational rad ius within a finite proper time in the experience of local free-falling
observers, in an (r,t) map their paths, given by } 1 dr + dt = O, appear to approach r = 2m
asymptotically as t --+ oc . Kruskal's ( 1960) prescription for removing the apparent
singularity is to replace (r,t) by new coordinates (u,v) which effectively 'straighten' the
paths of radial light rays to u ± v con st. Kruskal's coordinates are formally defined by
4m d ln(u ± v) = f - 1 dr ± dt
The early investigators took a mainly pragmatic view : the Schwarzschild

'singularity' is physically significant - i.e. actually exposed as part of the
exterior vacuum to which is appertains - only for bodies smaller than their
gravitational radius, and no such bodies were known. In his second paper,
Schwarzschild ( 19 16b) simply mentions without comment that this radius is
3 km for the sun and 10 - 2 8 cm for a 1 g mass. However, he also shows that a
static sphere of fluid with uniform energy density cannot, as a matter of
principle, be compressed all the way to its gravitational radius: the central
pressure will become infinite when the radius reaches i (2m).
Another reason of principle why r = 2m could apparently be ignored was
adduced by a 30-year-old Dutch scientist, Johannes Droste (19 16), a student
of H . A. Lorentz, in a paper communicated to the Amsterdam Academy at
the end of May. In this remarkable work, Droste independently derived the
Schwarzschild exterior solution and proceeded at once to a thorough and
elegant discussion of the orbits of particles and light rays in this field. He
observed that at r = 3m light rays would follow circular orbits. He also noted
that an infalling particle would take an infinite (coordinate) time to reach
r = 2m and concluded that 'we may, in studying its motion, disregard the
space r < 2m'. The theme of inaccessibility of r < 2m was often re-echoed :
There is a magic circle which no measurement can bring us inside. It is
not unnatural that we should picture something obstructing our closer
approach and say t hat a particle of matter is filling up the interior.
(Eddington, 1920)
Every gravitating particle has a ring-fence around it , which no other
body can penetrate.
(Whittaker, 1949)
Droste was the first to state clearly the operational significance of the
Schwarzschild radial coordinate : 'The circumference of a circle r = const . is
2nr . . . ; this shows how r can be measured'. Not everyone understood so
clearly that the operational meaning of the coordinates is (apart from global
identifications) entirely contained in the metric itself, and the role of general
covariance was a source of widespread confusion. How could the theory
lead to unique physical predictions if the coordinates could be changed at
will? Einstein himself had unhappily contributed to this confusion. In the
final months of 1959, while still groping his way toward the generally
covariant form of the field equations, he had calculated the perihelion
precession of Mercury using a special coordinate condition ( - g) 1 1 2 = 1 ,
which at that stage he still considered an essential element of the theory.
(This condition was also used by Schwarzschild, who did not have access to
234 W Israel
Einstein's concluding papers.) In 192 1 two noted scientists, the French

mathematician Paul Painleve ( 192 1 ) and the Swedish ophthalmologist and
Nobel laureate Allvar Gullstrand ( 1922),t independently set out to
demonstrate that Einstein's successful prediction was really an artefact of
this coordinate condition. They attempted to show that the answer for the
perihelion advance came out differently when calculated for the metric
ds 2 = dr 2 + r2(d82 + sin2 8 d</>2) + 2(2m/r) 1 ' 2 d r dt - (1 - 2m/r) dt2,
which is obtainable from the Schwarzschild metric by a simple change of the
time coordinate. Unwittingly, they had for the first time hit upon a form of
the metric in which 2m is manifestly non-singular. (In the 1930s essentially
the same form was rediscovered by Lemaitre and by Robertson, who
introduced exactly the same time coordinate.) Understandably, this passed
unnoticed at the time. Three years later, the opportunity slipped by again
when Eddington ( 1924c) , in a paper comparing Einstein's and Whitehead's
theories of relativity, found it convenient to introduce a time coordinate in
terms of which outgoing radial light rays in the Schwarzschild field travel
with unit speed. He thus obtained, without realizing it , another form non
singular at r = 2m that was rediscovered years later by Finkelstein ( 1958) .
Attempts to visualize how the geometry was distorted by the gravitational
field focused at first on the spatial slices of constant Schwarzschild time. By
ignoring one of the angular coordinates, a slice could be pictured as a surface
in a fictitious three-dimensional Euclidean space. It was shown by Flamm
( 19 16) that this surface is the upper half of a paraboloid obtained by
revolving a parabola whose axis is horizontal about a vertical axis, i .e. a
funnel-shaped surface terminating in a circle of circumference 2n(2m) at its
narrow end. Since the spatial geometry here is non-singular, there appeared
to be no obstacle to extending it to include the lower half of the paraboloid.t
This step looked even more plausible if one introduced an 'isotropic' radial
coordinate p by the substitution r = ( 1 + m/2p) 2p . Weyl ( 1 9 1 7) thus arrived
at a picture in which the region 0 < p < m/2 'interior' to the gravitational
t After a two-year public debate with C. W. Oseen and E. Kretschmann (e.g . 1923),
Gullstrand at length conceded . It was somewhat unfortunate that it was Gullstrand who
was chosen in 192 1 to prepare a confidential report on relativity for the use of the Nobel
committee (Pais , 1982). There was no physics award that year. The prize went to
Einstein in 1922 'for his services to theoretical physics and especially for his discovery of
the law of the photoelectric effect' . A valuable historical account of the Schwarzschild
solution up to 1923 has been published by Eisenstaedt ( 1982).
t It is, however, not clear from this argument that the 4-geometry is non-singular at the
throat. That the extended slice is a Cauchy surface for an initially regular vacuum
spacetime was not proved until 40 years later (Misner and Wheeler, 1957).
radius r = 2m fan s out to a second asymptotically flat space, linked to the

exterior space by a 'throat' or 'Einstein-Rosen ( 1935) bridge', as it was later
called. Since the extended three-dimensional slice is nested by 2-spheres
whose circumferential radius is never smaller than 2m, von Laue ( 1953)
inferred that 'points with r < 2m do not exist'. WeyI ( 19 17) deduced from the
express1 0n
( - g11) 1 1 2 = (p - m/2)/(p + m/2)
that 'cosmic' (i .e. coordinate) time and proper time would run in opposite
directions for a stationary observer in the 'interior' region . He concluded,
'Of course , only a segment of the solution that does not extend as far as the
singular sphere can actually be realized in Nature'.
The first explicit statement that r = 2m is not singular came from Lemaitre
( 1933) :
La singularite du champ de Schwarzschild est done une singularite
fictive, analogue a celle qui se presentait a } 'horizon du centre dans la
forrne originale de l'univers de Sitter.
He demonstrated this by exhibiting a coordinate system (essentially that
of Painleve and Gullstrand) in terms of which the metric at r = 2m was
manifestly regular.
Lemaitre's statement was buried as an incidental remark in the middle of
a long and somewhat inaccessible paper on cosmol ogy and it went largely
unnoticed. However, it did attract the attention of another cosmologist,
Howard Percy Robertson . Examining the orbits of radially moving test
particles in the Schwarzschild field Robertson noticed (see Robertson and
Noonan, 1968) a second point of resemblance between r = 2m and the
de Sitter horizon : although a particle dropped from any finite radius takes
an infinite coordinate time to reach r = 2m, the proper time measured by a
comoving observer is finite. The Schwarzschild 'singularity' is neither
singular nor inaccessible.
In 1939, Robertson gave a lecture on the Schwarzschild 'singularity' in
Toronto. (In the audience was the Professor of Applied Mathematics, J. L.
Synge, who was to retain vivid memories of the talk.) He also discussed the
problem with Einstein, who was sufficiently intrigued that in May he sent off
a paper whose declared aim was to examine 'whether it is possible to build
up a field containing such singularities with the help of actual gravitating
masses, or whether such regions with vanishing g 44 do not exist in cases
which have physical reality'.
The essential result of this investigation is a clear understanding as to
236 W Israel
why the 'Schwarzschild singularities' do not exist in physical reality . . . .

The 'Schwarzschild singularity' does not appear for the reason that
matter cannot be concentrated arbitrarily.
(Einstein, 1 939)
To bypass the uncertainties in the equation of state at high densities,
Einstein considered an idealized model whose microscopic dynamics he
could treat explicitly : in effect, a stationary relativistic 'globular cluster'
whose 'stars' follow circular Keplerian orbits. Einstein found that, for a
cluster of given mass there was a m i nimum size, 1! times larger than the
gravitational radius, below which the outermost 'stars' would have to travel
faster than light in order to stay in orbit. (This is really just the result of
Droste mentioned earlier.)
There is nothing wrong with this model, but in drawing such far-reaching
conclusions from it, Einstein slipped back into an oversight which had
vitiated his cosmological paper of twenty years before and which he
considered the biggest blunder of his life : he had again overlooked the
possibility of non-stationary matter configurations. That collapsing masses
which never reach a stationary condition are not only possible but
sometimes inevitable was not spelt out until two months later by
Oppenheimer· and Snyder.
Bergmann's standard text gives a fair impression of the general state of
understanding in the early 1940s :
The Schwarzschild field has a singular spherical surface at r = 2Km . . . .
Robertson has shown that, if a Schwarzschild field could be realized, a
test body . . . would take only a finite proper time to cross the
'Schwarzschild singularity' . . . and he has concluded that at least part
of the singular character of the surface r = 2Km must be attributed to the
choice of coordinate system.
In nature, mass is never sufficiently concentrated to permit a
Schwarzschild singularity to occur in empty space. Einstein considered
a system of many mass points . . . . The investigation showed that even
before the critical concentration of particles is reached, some of the
particles (the ones on the outside) begin to move with the velocity of
light . . . .
(Bergmann , 1942)
In 1949, John Lighton Synge, who had recently returned to Ireland as a
research professor at the Dublin Institute for Advanced Studies, decided
that it was time for a fresh assault on the problem. The purely formal
coordinate transformations which Lemaitre and Robertson had used to
'remove' the singularity and extend the spacetime left many unanswered
questions and obscurities . The original Schwarzschild metric is time
reversible and one would have expected this to carry over to any complete
extension. Yet the extensions of Lemaitre and Robertson were clearly not
reversible : they allowed only 'one-way traffic' through r = 2m. Moreover,
the old Flamm-Weyl-Einstein-Rosen picture of two asymptotically flat
spaces connected by a throat seemed quite incompatible with these
extensions. Synge saw this problem as an ideal proving ground for a cause he
had long championed : the geometric approach to general relativity.
Synge's attitude to the possible physical significance of such
investigations is made clear in the opening paragraphs of his paper.
Remarking that the Schwarzschild line element appears to be valid only for
r > 2m, he continues :
This limitation is not commonly regarded as serious, and certainly is
not so if the general theory of relativity is thought of solely as a
macroscopic theory to be applied to astronomical problems, for then
the singularity r = 2m is buried inside the body, i .e. outside the domain
of the field equations Rmn = 0. But if we accord to these equations an
importance comparable to that which we attach to Laplace's equation,
we can hardly remain satisfied by an appeal to the known sizes of
astronomical bodies. We have a right to ask whether the general theory
of relativity actually denies the existence of a gravitating particle, or
whether [the standard Schwarzschild metricJ may not in fact lead to
the field of a particle in spite of the apparent singularity at r = 2m.
(Synge, 1950)
He proceeds to exhibit the complete analytical extension of the
Schwarzschild manifold in terms of coordinates u, v with the property that
v + uconst are the paths of radial light rays , i .e. Kruskal-like coordinates .
Unfortunately, he obtained something extra. For r < 2m, Synge's
coordinates are inverse trigonometric functions of the standard Kruskal
coordinates. He was thus led to a picture of an infinite lattice of regions with
O � r � 2m in which test particles which have fallen through r = 2m will
oscillate forever. If truncated at the first of the geometrical singularities r = 0
- which would represent its maximal extension as a manifold - Synge's
extension is geometrically the same as Kruskal's. But, because of the
complexity of its analysis and the bizarre conclusions , its essential
significance was not understood.
In the view of many relativists the whole procedure of analytic extension
of spacetimes was a mathematical game without clearly defined rules (e.g.
238 W. Israel
Bel, 1969). Is the extended manifold necessarily unique (even aside from
topological ambiguities)? (Misner ( 1967) showed by a counter-example that
the answer is no.) Why insist on analytic continuation for hyperbolic
equations whose characteristic surfaces can admit discontinuities? At
what stage of the continuation does the mathematics take leave of physical
reality? The beginnings of a clear answer to these questions did not emerge
until the early 1 970s.
The decisive step in the unravelling of the extended Schwarzschild
manifold was taken by an amateur unaware of previous work in the field, the
noted plasma physicist Martin Kruskal. In the mid- 1950s Kruskal and a few
colleagues at Princeton formed a small study group to teach themselves
general relativity, using one of the current textbooks as a guide. Kruskal
noticed that the Schwarzschild metric near r = 2m closely resembles the flat
Minkowski metric expressed in uniformly accelerated ('Rindler')
coordinates. By simply following for the Schwarzschild case the path that
leads back from Rindler to Lorentz coordinates, he arrived at the well
known coordinates associated with his name.
Kruskal shelved his calculation when Wheeler, to whom he showed it,
reacted without special interest. A couple of years later (probably in 1958),
well aware riow of its importance, Wheeler reacquainted himself with
Kruskal's transformation and publicized it at the Royaumont conference on
general relativity in June 1959. By the end of 1959 the result had still not
appeared in print , though there were several references to its existence (e.g.
Finkelstein ( 1958) and Fronsdal (1959) mention it in notes added in proof).
Finally, in desperation , Wheeler was driven to writing up the work himself
in a brief paper for which author's credit was given to Kruskal ( 1960).
Meanwhile, others had independently tackled the problem without
knowledge of Kruskal's or other previous work . In a paper entitled 'Past
future asymmetry of the gravitational field of a point particle' David
Finkelstein ( 1958) at the Stevens Institute of Technology showed that
Schwarzschild's metric can be extended in two ways, using either retarded or
advanced time. At the time of submitting his manuscript, Finkelstein
believed that these were inequivalent extensions (rather than
complementary parts of a single complete extension) but was able to correct
this impression in a postscript added after Kruskal (whom he met at a
plasma physics conference) explained his work to him . (Finkelstein 's
extension remains of practical importance because it covers in a simple
explicit form as much of the exterior vacuum manifold as is needed for the
description of spherical collapse.) Results equivalent to Kruskal's were
derived in mid- 1959 by Christian Fronsdal , a particle physicist at CERN,

who followed the more complicated route of imbedding the Schwarzschild
manifold in a six-dimensional flat spacetime. At the same time the Adelaide
mathematician George Szekeres ( 1960) independently wrote down
Kruskal's transformation - he modestly credits his result 'essentially' to
Synge ( 1 950) - in an article which languished in obscurity in a Hungarian
journal of mathematics until 1966.
The scope of Szekere's article is actually much more general . It contains
the first careful definition (using Riemannian normal coordinates) and
classification of spacetime singularities. It also suggests for the first time -
and rejects - the possibility of an 'elliptic interpretation' of the extended
Schwarzschild manifold. Noting that each point of the original spacetime is
'represented twice' in the extension, he remarks,
Hence in order to obtain a manifold that represents physical reality, it
seems to be necessary to identify all pairs of opposite points (u, v) and
( - u, - v) ; . . . but . . . the Schwarzschild identification introduces an
artificial singularity at u = 0, v = 0 essentially of the same kind as the
singularity at the vertex of a cone obtained by identifying the points
(x, y) and ( - x, - y) of the Euclidean plane . . . it seems difficult to find
any physical justification for this identification process.
Since it is, indeed , 'hard to believe that every mass point should have the
effect of splitting the universe in two , thus necessitating a second copy of
ours' (Rindler, 1965), the elliptic interpretation was resurrected six years
later, with varying degrees of reservation, independently by a number of
authors (Souriau , 1965 ; Rindler, 196 5 ; Belinfante, 1966 ; Anderson and
Gautreau , 1966 ; Israel , 1966), but was soon abandoned . Quite apart from
the conical singularity there were obvious problems with causality which
proved to be insurmountable. t
7.8 Quasars and relativistic astrophysics (1951-72)

Last week Minkowski got the spectrum of the disputed object [Cygnus AJ
and promptly paid off the bet (one bottle of Scotch) which I had made
with him : the emission spectrum is indeed the outstanding feature
(redshift ""' 1 500 km/sec) and the high excitation indicated by the
t In the last few years, an analogous idea has been revived in a quantum context by
t'Hooft (e.g. 198 5 ; cf. Gibbons, 1986; Sanchez and Whiting, 1986).
240 W Israel
presence of the [NeV] lines! ! There the matter stands at present; the
collision case appears to be well established now.
(Baade, 1952)
We have this family of chaps, less than a second of arc across. We don't
know what to make of them.
(Hanbury Brown, 196 1)
Radar expertise (and surplus equipment) acquired· in World War II was
the stimulus that enabled British an9 Australian scientists to spearhead the
advance of radio astronomy in the post-war decade. This ushered in the
most eventful era in the history of astronomy since the time of Galileo. An
adequate account of these developments would require a book, and this
brief section cannot do more than sketchily trace one strand of this rich
tapestry. Numerous vivid accounts (many first-hand) that fill out the picture
are now available (e.g. J. L. Greenstein, 1963 , 1984; Robinson et al., 196 5 ;
Hey, 197 3 ; Bell Burnell, 1977 ; Sullivan, 1979; Hoyle, 198 1 ; Smith and
Lovell, 198 3 ; Ginzburg, 1984; G. Greenstein, 1 984; Hazard , 198 5 ; and
Blandford , this volume, Chapter 8).
.
In 1950 just one galaxy (Andromeda) was known to be a radio source, but
several dozen other discrete sources (generally much more powerful) had
been detected. Only three of these had been associated with visible objects :
one was coincident with the Crab nebula and two others (NGC4486 and
5 128) with 'nebulae' in Virgo (Messier 87) and Centaurus, whose status as
external galaxies was not to be established for another couple of years. The
prevailing consensus - Thomas Gold ( 195 1) was the only vocal dissenter
was that the discrete radio sources were dark objects, 'radio stars' (e.g. Ryle,
1950) within our galaxy. To place them outside the galaxy would have
ascribed to them an intrinsic radio power far exceeding Andromeda;
moreover, they were not correlated in position with prominent galaxies. It
even appeared that a local background of such discrete sources was actually
required to explain the long wavelength component of the continuum radio
emission from the M ilky Way. Kiepenheuer ( 1950) had already explained
this component as synchrotron emission from cosmic ray electrons
spiralling in the galactic magnetic field, and these ideas were actively
pursued by Ginzburg and others in the Soviet Union over the next few years.
But they did not sink in in Western astronomical circles until the mid- 1950s,
when measurements of the optical polarization of the Crab nebula
(proposed by I. M . Gordon) spectacularly vindicated Shklovsky's ( 1953)
idea that the synchrotron mechanism was at work in the Crab.
Astronomical opinion then rapidly evolved towards a consensus (Ryle,

1955) on the extragalactic character of the 'radio stars'.
A major breakthrough in this development had come in October 195 1
with the accurate positioning of the powerful radio source Cygnus A by F.
Graham Smith at Cambridge, using interferometer methods. By the
following summer Walter Baade at the Palomar Observatory identified it
with a faint dumbbell-shaped object which he interpreted as a pair of
colliding galaxies. Historically this was a turning point, the first
observational evidence that seemed to point clearly at the extragalactic
nature of the discrete sources and even to suggest a physical mechanism. But
no one claimed that all the sources could be explained in this way.
Regarding Messier 87 (now identified as an elliptical galaxy with a peculiar
jet), Baade and M inkowski ( 1954) remarked that 'no explanation can be
suggested as to why and how the pecularity in NGC4486 leads to strong
radio emission'.
Following a suggestion by Shklovsky ( 1955), it was established by Baade
in 1956 that the jet in M 87 is optically polarized. The true scale of the
energies involved at last began to emerge. According to synchrotron theory
(Burbidge, 1956), the energy stored in the jet in the form of relativistic
particles and magnetic field is equivalent to that released in 107 supernova
explosions.
Even as an explanation for specific sources, the collision hypothesis
remained in vogue for only a few years. (It seemed to be difficult, for
instance, to obtain a high efficiency of conversion from low-grade collisional
energy into relativistic particles (Burbidge, 1958).) By the turn of the decade
(see Struve, 1960) only the original proponents of the idea were still ready to
defend it (Minkowski, 196 1).t It was slowly ousted from favour by the
hypothesis - stemming from Viktor A. Ambartsumian ( 1958) and promoted
vigorously by the Burbidges - of violent outbursts in the nuclei of galaxies,
due to causes unknown.
In a forceful and provocative address to the 1958 Solvay Congress,
Ambartsumian had marshalled an array of empirical evidence, from the
existence of polarized jets to the circumstance that clusters of galaxies
t Mounting evidence of a correlation between quasar-like activity and membership of
close interacting pairs is currently reviving the fortunes of the collision hypothesis, in
particular for Cygnus A (Thompson, 1984) and similar galaxies (Gaskell, 1985;
Begelman et al., 1980). There is an interesting blend of the collision and accretion
hypotheses in some recent quasar theories, which suggest that accretion may be fuelled
through entanglement of the material in the overlapping nuclei (each containing a black
hole) belonging to a giant elliptical and a satellite dwarf galaxy.
242 W Israel
appeared to be gravitationally unbound , all of which pointed, in his opinion,

to the inference that galactic nuclei were the seal of explosive events that
gave birth to new galaxies :
Apparently we must reject the idea that the nucleus of a galaxy is
composed of common stars alone. We must admit that highly massive
bodies are members of the nucleus which are capable not only of
splitting into parts that move away at a great velocity but also of
ejecting condensations of matter containing a mass many times
exceeding that of the sun . . . .
. . . the primary nucleus of a galaxy for reasons unknown to us is
divided into separate parts which give birth to independent galaxies
that become components of the system. In this case the process of
division must take place in [aJ small volume with a diameter measuring
some parsecs or tens of parsecs . . .
Radio galaxies Perseus A and Cygnus A are systems in which
division of nuclei has taken place but the separation of galaxies is not
yet complete.
The hypothesis of large-scale ejection of matter from galactic nuclei had
already been invoked 30 years before by Jeans in an attempt to explain spiral
structure :
The type of conjecture which presents itself, somewhat insistently, is
that the centres of the nebulae are of the nature of 'singular points' , at
which matter is poured into our universe from some other, and entirely
extraneous, spatial dimension, so that, to a denizen of our universe,
they appear as points at which matter is being continually created .
(Jeans, 1928)
These sentiments foreshadow the 'white hole' quasar theories of the mid-
1960s (Novikov, 1964 ; Ne'eman, 1965).
With the theory in this unsettled state, the observers were on the threshold
of remarkable new developments. Optical counterparts of many radio
sources had been found by 1960 and identified as distant galaxies.
Minkowski found that one object , the source 3C295, was a galaxy in Bootes
which had the unprecedented redshift of 0.46. The interest of the
cosmologists was aroused . Towards the end of 1960, Allan Sandage at
Palomar and a group of collaborators - the radio astronomers Thomas A.
Matthews and T. G. Bolton and the spectroscopists Jesse Greenstein and
Guido Munc� - initiated a careful study of three unidentified sources whose
large radio flux and small angular size (less than a few seconds of arc
Dark stars: the evolution of an idea · 243
according to radio interferometer measurements at Jodrell Bank)

encou raged the hope that they might be galaxies of very large redshift.
Sandage presented a preliminary report on this study in an unscheduled
talk at the New York meeting of the American Astronomical Society (AAS)
at the end of December 1960.
The first object studied was 3C48 . . . . A direct plate was taken on
September 26, 1960 with every expectation of finding a distant cluster of
galaxies , but measurement of the plate gave the unexpected result that
the only object lying within the error rectangle of the radio position was
one which appeared to be stellar. The stellar object was associated with
an exceedingly faint wisp of nebulosity . . . .
(Matthews and Sandage, 1963)
The contemporary report on the AAS meeting in Sky and Telescope ( 196 1 )
gave this account :
Spectrograms taken by Jesse L. Greenstein and Guido Munch of
Caltech , and by D r Sandage, show a combination of strong emission
and absorption lines unlike that of any other star known . . . . The
spectrum shows no hydrogen , the main constituent of all normal stars.
Since the distance of 3C48 is unknown, there is a remote possibility
that it may be a very distant galaxy of stars ; but there is general
agreement among the astronomers concerned that it is a relatively
nearby star with most peculiar properties. It could be a supernova
remnant.
The spectrum of 3C48 remained an enigma for the next two years,
although the possibility of 'a big redshift on the spectrum' was raised
informally more than once (Hoyle, 198 1). Matthews and Sandage observed
variations in optical brightness of up to 30 per cent in less than a year,
seemingly irrefu table evidence that 3C48 was not a galaxy . As late as
February 1963 Bu rbidge was speculating on the possibility that 3C48 and
the other 'radio stars' were neutron stars and 'that the spectral featu res
might arise from elements which are normally rare and that there may be
considerable gravitational redshifts' .
However, the real breakthrough came from a different direction . A rare
and fortunate series of lunar occultations of the powerful source 3C273
which occu rred in April, August and October, 1962, were observed by
Hazard, Mackey and Shimmins ( 1963) with the 2 10-foot steerable dish at
Parkes, Australia. They revealed a double radio structure and gave the
position with the unprecented accuracy of less than a second of arc. An
244 W Israel
optical search by the Caltech astronomer Maarten Schmidt ( 1 963) (to whom
Hazard had sent his result at the suggestion of John Bolton), now showed
that
The only objects seen on a 200-inch plate near . . . the radio source
3C273 . . . are a star of about thirteenth magnitude and a faint wisp or
jet. . . .
Spectra of the star were taken with the prime-focus spectrograph of
the 200-in . telescope . . . . A redshift dA./A.0 of 0. 1 58 allows identification
of four emission bands as Balmer lines . . . .
Greenstein ( 1963) immediately re-examined his 196 1 plates of 3C48 . He and
Matthews were now able to identify six lines by applying a redshift of 37 per
cent.
All these revelations appeared back-to-back in the March 16 issue of
Nature along with a theoretical paper by Hoyle and Fowler ( 1963),
proposing a collapsing superstar hypothesis, which concluded :
Our present opinion is that only through the contraction of a mass of
107- 108 M 0 to the relativity limit can the energies of the strongest
sources be obtained.
An earlier suggestion that gravitational energy released in cont raction
might account for the radio sources had come from Ginzburg ( 196 1) :
. . the gravitational contraction of a galaxy (or of its central region),
.
which is accompanied by the appearance of motions and

inhomogeneities in the gas (instability) and , subsequently star
formation , must also lead to the generation of cosmic rays. It is easy to
see the energy required (up to 106 1 erg in the case of Cygnus A) may
have a gravitational origin.
In June 1963 , Ivor Robinson ( 1965), Alfred Schild, Engelbert Schucking
and Peter Bergmann sent out an invitation to more than 300 scientists which
read in part :
The intriguing new discoverie� and the theory put forward by Hoyle
and Fowler open up the discussion of a wealth of exciting questions.
Among the problems raised are the following :
(a) The astronomers observed some unusual objects connected with
radio sources. Are these the debris of a gravitational implosion?
(b) By what machinery is gravitational energy converted into radio
waves?
(c) Does gravitational collapse lead, on our present assumptions, to
indefinite contraction and a singularity in space time?
(d) If so , how must we change our theoretical assumption to avoid

this catastrophe?
. . . After checking with some of the protagonists, we have come to the
conclusion that a conference late this year at the Southwest Center for
Advanced Studies in Dallas, Texas, might be well timed.
The First Texas Symposium on Relativistic Astrophysics took place in a
downtown Dallas hotel from 1 6 to 18 December, 1963 (Sky and Telescope,
1964). It was an event which no one who participated is likely to forget. The
opening mood was sombre as the scientists were welcomed by Governor
John Connally, his arm in a cast and sling from the tragedy which had
happened a few blocks away on November 22. But as the epic discoveries
unfolded the rising excitement could not be contained . Although my own
graduate research had been in general relativity , this was the first time I
heard the work of Oppenheimer and Snyder being discussed . No doubt
many others were in the same situation . I recall an ebullient Roy Kerr ( 1963)
treating me in my hotel room to a preview of his talk on the topological
structure of his recently published metric : 'Pass through this magic ring and
- presto ! - you're in a completely different universe where radius and mass
are negative! ' This was heady stuff, but it seemed to bear no relation to
physics. None the less, these were heartening words from Thomas Gold at
his after-dinner speech :
This, of course, is a historic meeting. It will be remembered as the
meeting where these great new astronomical discoveries were first
discussed . It will also be remembered for the display there of strong men
wrestling with even stronger facts.
It was, I believe, chiefly Hoyle's genius which produced the extremely
attractive idea that here we have a case that allowed one to suggest that
the relativists with their sophisticated work were not only magnificent
cultural ornaments but might actually be useful to science! Everyone is
pleased : the relativists who feel they are being appreciated, who are
suddenly experts in a field they hardly knew existed ; the astrophysicists
for having enlarged their domain, their empire, by the annexation of
another subject, general relativity . It is all very pleasing, so let us all
hope that it is right. What a shame it would be if we had to go and
dismiss all the relativists again.
At one of the opening sessions Greenstein addressed the possibility that
3C48 might be a comparatively nearby object with a redshift of gravitational
origin . It is reconcilable, he concluded , with the spectral features (sharp
emission lines, presence of forbidden lines) if the source is nearer than the
246 W Israel
moon, but astronomers do not seem to need such . an object to account for
planetary motions!
Harlan Smith reported on a study of old Harvard plates which revealed
optical variations in 3C273 on timescales of years and months, with even
some indications of flashes in which the brightness doubled in less than ten
days. It appeared that the primary source of energy could not be more than a
light-week in diameter.
Most of the Soviet scientists invited were unable to attend, but they sent
preprints which circulated freely. An extraordinary one, by Zel'dovich and
Novikov (1964), envisaged a supermassive object powered by accretion
(Zel'dovich, 1964; Salpeter, 1964), and argued that its mass must be at least
10 8 M 0 if the observed luminosity is not to exceed the Eddington limit. The
strongest source could be adequately fuelled by accretion of just a few solar
masses per year. Ginzburg (1964), Ozernoy and Kardashev ( 1964) stressed
the role of magnetic fields, which would be strongly amplified by flux
freezing when a compact object forms in a collapse. These ideas paved the
way for the theoretical understanding of pulsars a few years later.
The basic difficulty with gravitational collapse as a power source was
emphasized by Freeman Dyson in one of the concluding summaries. The
timescale for · collapse is only about a day; the lifetime of the sources,
inferred , for example, from the length of the jets, is at least 106.- 107 years.
Attempts over the next couple of years to meet this difficulty head-on by
replacing the Oppenheimer-Snyder scenario by an oscillatory picture (e .g.
Gertsenshtein, 1967) came to nothing. Modifying the Einstein field
equations with a negative-energy cosmological - or 'C-term ' - (Hoyle and
Narlikar, 1964 ; Faulkner et al. , 1964) introduces repulsive gravitational
forces which can lead to a bounce at small radius, and a pulsation as seen by
a comoving observer. On the other hand, it seemed obvious (Zel'dovich and
Novikov, 1965) that an external observer could never see the bouncing
object re-emerge from the event horizon r = 2m that it has not yet entered in
his remote future! The elliptic reinterpretation of the extended
Schwarzschild spacetime predicts oscillatory behaviour for infalling test
particles at the price of relinquishing ordinary ideas of causality within a
sphere slightly larger than the gravitational radius, which , it could be - and
was - argued, might lie beyond the range of conscious intervention (Israel ,
1966). Even so , it proved impossible to devise a consistent dynamical picture
of a gravitating source which revisits the same Cauchy su rface.
There were, of course, many other hypotheses about quasars and active
galactic nuclei : chain reactions of supernova explosions in distant galaxies,
matter-antimatter annihilation, white holes, quasi-local hypotheses in
which the quasars were massive objects ejected at relativistic speeds from the
nucleus of our own galaxy, and so on. In retrospect it seems clear that many
of these ideas were over-reactions to phenomena whose power (as distinct
from total energy) requirements did not call for such extraordinary physics .
Only one or two seem to have stood the test of time : the model suggested by
Lynden-Bell ( 1969) for the more quiescent sources in which a super-massive
black hole is fed by an accretion disc,t and the 'magnetoid' or 'spinar' model
of Ginzburg and Ozernoy ( 1 977) , Morrison and others , in which the energy
source is a magnetized supermassive disc stabilized by rotation. As seen
from a present-day vantage point the situation in the mid- 1 960s has been
well summarized by Rees ( 1 984) :
There has been progress towards a consensus in that some bizarre ideas
that could be seriously discussed a decade ago have been discarded . But
if we compare present ideas with the most insightful proposals
advanced when quasars were first discovered 20 years ago (such
proposals being selected , of course , with benefit of hindsight) , progress
indeed seems meager. It is especially instructive to read Zel' dovich and
Novikov's ( 1964) paper entitled 'The mass of quasi-stellar objects' . In
this paper, on the basis of early data on 3C273 , they conjectured the
following : (a) Radiation pressure perhaps balances gravity, so the
central mass is - 108 M 0 . (b) For a likely efficiency of 10 % the
accretion rate would be 3 M 0 yr- 1 . (c) The radiation would come from
an effective 'photosphere' at a radius of - 2 x 10 1 5 cm (i .e. � r9 ), outside
of which line opacity would cause radiation to drive a wind. (d) The
accretion may be self-regulatory, with a characteristic time scale of
,..., 3 yr. These suggestions accord with the ideas that remain popular
today, and we cannot yet make any firmly based statements that are
more specific.
By good fo rtune the fourth 'Texas' symposium in Dallas in December
1968 was perfectly timed to catch the excitement of the unravelling of the
pulsar mystery. It was at the meeting itself that most delegates learned of the
Crab pulsar's deceleration measured at Arecibo . Rotational energy was
being lost at a rate which exactly accounted for the energy radiated (mainly
in X-rays) by the nebula! At one stroke Gold's ( 1968) model of pulsars was
vindicated and the existence of neutron stars established beyond reasonable
doubt. To borrow a well-worn phrase, seldom in the history of astronomy
has so bizarre a concept been so firmly established in so short a time,
although the 'central star' in the Crab, whose spectrum showed a blank, had
248 W Israel
for 25 years been suspected of being the seat of mysterious, highly energetic
processes (Minkowski , 1942 ; Hoyle, 1955). A hint of the true source of the
Crab's radiation had come from Wheeler ( 1 966) in a review of 'superdense
stars' :
. . . several workers have looked to the residual neutron star itself as a
device to power this radiation , either through its heat energy or
through its energy of vibration . . . . (Energy of rotation appears not yet
to have been investigated as a source of power. Presumably this
mechanism can only be effective - if then - when the magnetic field of
the residual neutron star is well coupled to the surrounding ion clouds.)
Spinning neutron stars with oblique magnetic axes had been studied by
Pacini ( 1 967) several months before Hewish's ( 1 968) group announced their
discovery. Yet no one managed to predict that such objects would pulse , and
the exact mechanism remains a matter of controversy to this day.
X-ray astronomy owes its inception in the late 1940s to another legacy of
World War I I : a small stock of V2 rockets captured from the Germans . X
rays from outside the solar system were first detected in 1962 . In 1963, taking
advantage of a lunar occultation , Herbert Friedman and his colleagues at
the Naval Research Laboratory identified the Crab Nebula as a diffuse X
ray source, and, in 1964, they identified the irregularly variable source
Cygnus X- 1 . By 197 1 , after it had been located to within a few minutes of arc
by Uhuru , Cygnus X- 1 was linked with a variable radio source which was
then more accurately pinpointed by radio astronomers in Green Bank and
Holland. It was finally identified with the one-line optical binary HOE
226868 by L. Webster and P. Murdin at the Royal Greenwich Observatory
and by C . T. Bolton ( 1972) at the David Dunlap Observatory (Toronto).
The mass of -6 M 0 estimated for the unseen component made it the first
durable observational candidate for a black hole, and it remained the only
one for more than a decade.t
7.9 Non-spherical collapse : from frozen star to black holet

(1964-71 )
. . . i t becomes dimmer millisecond b y millisecond, and i n less than a
second is too dark to see. What was once the core of a star is no longer
t In December 1982 , a team at the Dominion Astrophysical Observatory found evidence
for a massive invisible component in the extragalactic X-ray binary LMC X-3 (Cowley et
al., 1983).
! The reader is cautioned that this section is little more than a personal reminiscence.
Review articles by Carter ( 1973) and Tipler, Clarke and Ellis ( 1980) contain objective
historical accounts, particularly of developments in the 1 960s .
Fig. 7 .1. Is there a black hole in Cygnus X-1? A bet between Stephen
.
Hawking and Kip Thorne made at Caltech in December 1974, on which
neither side has yet collected.
C.Uia.'\&o." S�pk"' UG.�R�

lo.a su� � e:-.1e �.,,�t\.f c.ft.
Gr�e �� ct 4iGu1e
Uo6 c o..nd dasi ,.._, o.f'\ C:n1uf"'Q.r\�
p>f,c.". ,Q.f'\J � tu. te� � �u
� 0.c.t d.�e.�c.icl'f wi.tJ.. Of.rJ- a.n
'nsuro..f\.ce fid•c.'(
,
Th.ere.lore. C:> & \J- re sot '1t-Gl. th.J.
$1�ph·" 1-4�1\c;.. Cj .e.&s ' 1t1M.'•
a c#ac.n.'p+iO"' t. ��,.I-house. - ..� �..; ..t
...
Kcf. Tfert.,_ � w...7"" of. o.. t+ -'f�

Sb�sc,re.pt'eo"· � -� c:1� , tha..,. . ..
C"f � )( I oto., nd-' C..Ot\.ha:, n. CL

I.lo.ck �ott. of �a.�s a,,kw. th.e.
Ck�n-al � kt� e(,..�t .
•� I(�.!.�· -
'f:i � . �·�;;7,1 . r·k ':- -'. " .1 . ""·.'
11'.I • r
.-.I "'1"":
... .
,.. �·� ... �
. .
• ., ,,,. ! � . .'J.
.
u�... w
•
,�.... ...��
250 W Israel
visible. The core like the Cheshire cat fades from view. One leaves
behind only its grin, the other, only its gravitational attraction . . . .
Moreover light and particles incident from the outside emerge and go
down the black hole only to add to its mass and increase its
gravitational attraction.
(Wheeler, 1968)
Black hole research is a good example of the importance of pictures and
phrases. Before the mid- 1960s the object we now call a 'black hole' was
referred to in the English literature as a 'collapsed star' and in the
Russian literature as a 'frozen star'. The corresponding mental picture,
based on stellar collapse as viewed in Schwarzschild coordinates was
one of a collapsing star that contracts more and more rapidly as the
grip of gravity gets stronger and stronger, the contraction then slowing
because of a growing gravitational redshift and ultimately freezing to a
halt at an 'infinite-redshift surface' (Sch warzschild radius), there to
hover for all eternity. Of course, from the work of Oppenheimer and
Snyder ( 1939) we were aware of an alternate viewpoint, that of an
observer on the surface of the collapsing star who sees no freezing but
instead experiences collapse to a singularity in a painfully short time.
But because nothing inside the infinite-redshift surface can ever
influence the external universe, that 'comoving viewpoint' seemed
irrelevant for astrophysics. Thus astrophysical theorizing in the early
1960s (e.g . Zel'dovich and Novikov, 1964, 1965) was dominated by the
'frozen-star viewpoint'. As long as this viewpoint prevailed , physicists
failed to realize that black holes can be dynamical , evolving, energy
storing and energy-releasing objects.
(Thorne et al., 1986)
In 1963 , all at once putative experts in a field they had left untilled for a
quarter of a century, the relativists scurried to make up for lost time. Their
success exceeded all expectations. Within a few years, understanding of
gravitational collapse progressed from its inchoate beginnings to a
sophisticated discipline comparable in elegance, rigour and generality to
thermodynamics, a subject which it turned out unexpectedly to resemble.
The array of problems was formidable. For the first time, the full non
linearity of the Einstein field equations had to be confronted for a generic
situation. Astrophysics insistently posed the question : what is the evolution
and endstate of a generic collapse with all complications included : pressure,
rotation , magnetization , etc? For example, when a star collapses toward its
gravitational radius , one expects its magnetic field to become strongly
amplified by flux-freezing as its surface area contracts. Could the final

collapsed object thus retain a permanent magnetosphere sufficiently
powerful to accelerate particles in a surrounding plasma to the relativistic
energies that one observes in the Crab nebula and quasar-like sources?
(Ginzburg , 1964 ; Ginzburg and Ozernoy, 1 965.)
To guide the way toward answers to such questions on the nature of a
generic collapse only one analytical solution was available. This was the
highly idealized and special Oppenheimer-Snyder scenario for the collapse
of a spherical cloud of dust, in which the exterior field was represented by the
Schwarzschild metric. The stationary axially symmetric vacuum metric
which had recently been found by Kerr ( 1963) contained an event horizon
(similar to r = 2m) and thus conceivably represented an exterior field for the
endstate of some spinning collapse, though again (presumably) of a very
special kind, since the solution involved only the two minimal parameters,
mass and angular momentum. Both solutions had bizarre features which
made it debatable whether they could be at all representative of a generic
collapse.
In the first instance, both metrics are marked by singularities of infinite
cu rvature. In the Schwarzschild case this singularity (at the centre, r = O) is
spacelike, thus obstructing attempts to follow the evolution past the
moment of maximal compression. On the other hand, it seemed plausible to
argue, and was generally believed in the early 1960s (e.g. Lindquist and
Wheeler, 1957; Zel'dovich , 1 965) , that this central singularity was entirely
artificial , an artefact of the special initial conditions (spherical symmetry)
imposed on the solution, which had the effect of aiming all the particles at
one point. Moreover, this view received support from a detailed analysis by
Lifshitz and Khalatnikov ( 196 1 , 1 963) which concluded that singularities
are not a generic feature of solutions of the Einstein field equations :t
Therefore the results that have been set forth exclude the possibility of
the existence of a singularity in the future and mean that the contraction
t Specifically, the argument was that a generic solution involves eight arbitrary functions
of the spatial coordinates (six components of the spatial metric tensor and their six time
derivatives less four degrees of freedom due to arbitrariness of the spacetime
coordinates), whereas a singular solution appeared to allow only seven arbitrary
functions. The conflict with the singularity theorems of Hawking and Penrose ( 1969) was
later traced to a subtle point (Khalatnikov and Lifshitz, 1970 ; Misner, 1969) : the generic
solution does not have an analytic expansion near a singularity as Lifshitz and
Khalatnikov had originally assumed , but exhibits a complicated stochastic ('mixmaster')
behaviour.
252 W. Israel
of the world (if it should occur in general) should in final analysis

alternate with its expansion .
(Lifshitz et al. , 196 1 )
The Kerr metric, analytically extended beyond its horizon , revealed even
more alarming idiosyncrasies. The region within the inner horizon admits
closed timelike cu rves, implying an irremediable breakdown of causality in
this region. If the analytically completed solution was indeed relevant to
collapse rather than just a mathematical curiosity, the outlook was grim :
All these things suggest that the breakdown in general relativity may be
of a global rather than (or as well as) of a local nature, in which case it is
very serious indeed. If this tu rns out to be the case then one will not be
able to expect to cure the trouble by minor modifications significant
only in regions of high curvature, so that the whole theory might have
to be abandoned , or at least drastically reformulated.
(Carter, 1968)
The possibility that the available, very special models of collapse might be
totally misleading as a basis for induction led to some d rastic proposals for
modifying the Oppenheimer-Snyder scenario or 'rationalizing' the region
within the horizon , including the idea of non-analytic continuation (Komar,
1 965) and others already mentioned in Section 7.7.
An essential stabilizing influence during this period was the appearance of
a two-part review article of Zel'dovich and Novikov ( 1964--65), which
surveyed and virtually defined the new field of 'relativistic astrophysics'. A
mine of sound physical judgement and good sense whose incisive analyses
drove straight to the heart of every problem , 'often doubtful but rarely
wrong' (Wolfendale, 1986), it became the bible on gravitational collapse
until the end of the decade, when it was succeeded by the 'New Testament' of
Roger Penrose ( 1969). There could be no doubt, the authors stoutly insisted,
that for an external observer the Oppenheimer-Snyder picture of collapse to
a 'frozen' state must be generically correct, at least for small deviations from
spherical symmetry and small angular velocities. As to what happens inside
the event horizon (the ' T-region') , they were again characteristically
forthright :
We still have no complete answer to this question. We can only state the
following. According to the derivations of Lifshitz, Sudakov and
Khalatnikov the matter cannot contract to infinite density. As we have
seen above, the star cannot expand again , even in an asymmetrical
manner, so as to come out from under the Schwarzschild sphere into a
region which can be accessible to an external observer. It is possible
that the development of the asymmetry leads to a stronger variation of

the geometry of space-time inside the T-region or even to chang e in the
topology . At any rate, no matter what occurs inside the T-region, this
will never be manifest in the spacetime region outside the
Schwarzschild sphere and the external observer will never learn
about it .
Soon after the appearance of this article, it was recognized (N ovikov,
1966 ; de la Cruz and Israel , 1967) that a spherical star actually could
rebound from inside the Schwarzschild sphere if it was endowed with an
arbitrarily small amount of charge or by dint of such devices as the
introduction of the repulsive, negative energy 'C-field' of Hoyle and
N arlikar. The star would then re-emerge from a past horizon of the
analytically extended manifold . To preserve causality, one was then forced
to assume that this past horizon was part of a different universe ! To many
people this picture seemed at least as fantastic as the alternative of
irreversible collapse to zero volume, although subscribers to white hole
theories of quasars may have found it somewhat less difficult to swallow.
However, all bets were off concerning the possibility of a non-singular
outcome to a collapse after the publication of the theorem by Roger Penrose
( 1965), which has claims to be considered the most influential development
in general relativity in the 50 years since Einstein . founded the theory.
Penrose demonstrated that singularities were, after ai l , a generic feature of
gravitational collapse and must appear soon after the formation of a trapped
surface (a surface from which light cannot escape outwards). Although the
theorem could offer no information about the nature of the singularity, there
was a widespread belief that it would have an all-enveloping spacelike
character, as in the Schwarzschild case, thus obstructing further
development of the solution.
Penrose's paper was just as important for what it initiated as for what it
accomplished . Powerful global techniques exploiting the causal structure of
spacetime had been introduced into the theory for the first time. In the hands
of Hawking and Penrose ( 1969) , with essential input from Robert Geroch
(e.g. 1970), these techniques were developed to yield far-reaching results on
the occurrence of singularities in cosmology. Their application to
gravitational collapse culminated in the elegant results of Hawking and Ellis
( 1973) on the global properties of event horizons and , in particular,
Hawking's ( 197 1) proof of 'the second law of black hole mechanics', which
states that the area of the event horizon can never decrease.
An indispensable aid to clarity was the ingenious 'conformal diagram'
254 W Israel
invented by Penrose and extended to spacetimes with event horizons by

Carter ( 1966). This is essentially a Kruskal diagram compactified so as to
bring infinity and all horizons onto finite parts of the map. With its aid one
could survey at a glance the global structure of the most complicated
spacetimes and it quickly became a standard tool of the trade.
The rationale for analytically extending spacetime manifolds, and the
physical status of the more bizarre extensions, like that of the Kerr metric,
could now be clarified. Relativists began to understand that the natural
guiding principle was · not indiscriminate analytic continuation , but rather
the maximal Cauchy development of generic regular initial data. In other
words, one should be able to trust a solution of the field equations, evolving
from regular initial data, up to the point where it, or any perturbation of it
obtained by generically perturbing the initial data, becomes singular . This
principle identified the natural boundaries of an extension as the compact
sheets of Cauchy horizons, in particular the ingoing sheets of inner event
horizons ; these act as infinite blue-shift surfaces which magnify
perturbations. Thus, as was first emphasized by Simpson and Penrose
( 1973), the oscillatory collapse of a charged sphere is an unstable
idiosyncrasy of exact spherical symmetry ; a generic collapse terminates at
the inner horizon, which is generically singular.
However, we must return to the era preceding this age of enlightenment in
order to follow another line of development . The jubilee year of general
relativity saw the appearance of another major and richly suggestive paper,
by Doroshkevich, Zel'dovich and Novikov ( 1965) (DZN), which launched a
broad attack on the problem of non-spherical collapse.
The authors' first aim was to show that the Oppenheimer-Snyder
scenario is stable, i.e. that collapse with small deviations from spherical
symmetry proceeds qualitatively like spherical collapse and that an event
horizon still forms. Their argument, in essence, was that, although
asymmetries might grow without bound as the collapse proceeds to zero
volume, the resulting blowup of the field , propagating causally along
characteristics of the unperturbed spacetime, could never reach the horizon
itself. Although this argument was plausible, it would have been impossible
to make it rigorous with the mathematical techniques then (or now)
available; these allow one to predict the behaviour of a solution from initial
data for a finite time interval only, whereas the horizon must persist for an
infinite time. This question , which remains today the outstanding unsolved
mathematical problem of the field, was later codified as the 'cosmic
censorship hypothesis' by Penrose ( 1969).
DZN next add ressed the question of the nature of the final 'frozen' state,
in which the external field becomes asymptotically stationary. Examination
of special cases indicated that static fields with multipole moments become
singular at the event horizon. They went on to argue as follows :
An analysis of the static solution outside the body shows that the
deviation from the spherical solution , which is caused by a change in
the sou rce of the field , leads to the appearance of true singularities of
spacetime on the Schwarzschild surface g00 = 0. On the other hand , in
the comoving system of a contracting body with small initial deviations
from sphericity in the density distribution, the Schwarzschild surface is
in no way specially distinguished and is not accompanied by the
appearance of true singularities either in the metric or in the density. A
comparison of these results leads to the conclusion that the quadrupole
and higher moments of the external gravitational field attenuate during
the relativistic stages of collapse. t
This conclusion was highly intriguing but puzzling, since no dynamical
explanation was evident or, indeed, offered for the decay of the multipole
moments. One possibility, suggested later, was that it was an apparent
effect , seen from the outside, due to 'Lorentz contraction of the bulges on the
star's surface' (Thorne, 1967) . DZN were willing to commit themselves only
on what the cause was not :
The change in the multipole moments during the course of contraction
of the body should be accompanied by radiation of gravitational waves ,
but the energy carried by this radiation is small. The radiation of waves
is a consequence of the change of the multipole moments and should
not be regarded as the cause of their total damping. We note that in
Newtonian theory, the moments also vary du ring the course of the
compression of the body, but for finite body dimensions they are finite.
In Einstein's theory, a relativistic damping is superimposed on this
change in the moments of the external field , due to the change in the
dimensions of the contracting body.
The external observer 'sees' (for example, with the aid of neutrino
and antineutrino radiation) in the ultimate cooled state the finite
nonsphericity of the distribution of masses in the sources of the field .
However, this nonsphericity is not at all manifest in the external field .
t Ginzburg and Ozernoy ( 1965) had noted a few months earlier that, for an external
observer, the effective magnetic moment of a magnetized spherical star contracting
quasistatically toward its gravitational radius appears to attenuate like t - 1 .
256 W Israel
The authors now proceeded to consider collapse with rotation . Here,

unfortunately, they were diverted from an anticipation �f the 'no hair
conjecture' by the fact that the only form of Kerr's metric then known
contained 'non-removable off-diagonal components gµv in addition to g 0 3 ' .
They interpreted this to mean that any material source of the Kerr metric
must have meridional circulation as well as spin. They concluded that there
must be a great variety of endstates for spinning collapse, amongst which the
Kerr fields were merely a special class. In fact, the unwelcome off-diagonal
elements are actually removable, as Boyer and Lindquist ( 1 967) were to
show by an explicit coordinate transformation , and Papapetrou ( 1966) by a
general theorem, a year or two later.
I recall being struck by this paper when I came across the English
translation in the physics library at the Technion (where I was a visitor in the
spring of 1966) and my baffiement at some of its passages. One part that I
could immediately connect with was the discussion of static perturbations of
the Schwarzschild metric. I had previously done some work on this (Israel
and Khan , 1964), inspired by a preprint sent to me in 1963 by a former
University of Alberta student, Lawrence Mysak , who had gone to Adelaide
to work with George Szekeres (Mysak and Szekeres, 1966). t Our study had
shown rather generally that static event horizons remain non-singular but
become tidally deformed by the presence of external bodies. If, therefore, it
was assumed that no external bodies were present , and if the DZN
contention about the effect of multipole moments was indeed correct (as I
suspected), then it should be possible to prove that every non-singular static
event horizon must be spherical.
This was not followed up immediately because I was still caught up with
the elliptic interpretation (Section 7. 7) and possibilities for oscillatory
collapse. In the fall of 1966, I began a sabbatical leave at the Dublin Institute
for Advanced Studies, accompanied by my student Vicente de la Cruz, who,
sensibly, did not share my brief infatuation with the elliptic interpretation .
Toward the end of that year, confronted by the preposterous need t o
postulate a new universe every time a charged sphere bounced from under
its gravitational radius (de la Cruz and Israel, 1967), and disgusted with the
failure of the elliptic interpretation to provide a consistent way out of the
mess, I turned to the possibility that the picture of an exactly spherical
bounce might be generically unstable. This idea was good , but in my
thinking I placed the locus of instability at the event horizon instead of at the
t Because of difficulties with referees, this paper had to wait three years for publication.
infinite-blueshift sheet of the inner horizon where it belonged. It then

occu rred to me that the property that only spherical static horizons were
non-singular could be taken as indicating that the formation of horizons in
gravitational collapse is a non-generic and unstable feature - in a realistic
collapse no horizon should form at all .
At any rate i t seemed worthwhile a t this point to put the sphericity of
static horizons on a firm footing. In a short time I was able to prove a
theorem to this effect, using some identities of a generalized Green's type
which the static field equations provided very obligingly - and rather
surprisingly, since they are non-linear. The general method of the proof
appeared to be naturally extendable to the stationary case, and the very
similar algebraic structures of the Schwarzschild and Kerr metrics made it a
good bet at this stage that one should be able (with considerably more
labour!) to show that the Kerr horizons were the unique stationary
horizons.
This theorem , and my attempt to interpret it as indicating instability of
horizons , formed part of a seminar on 'gravitational bounce' which I
presented at King's College, London , in February 1967. It was part of a
double bill (J . P. Vigier was the other speaker) and there was a large turnout,
with Bondi , Pirani, Carter and Misner (then on sabbatical leave in
Cambridge) among the people in the front row.
My arguments for ho rizon instability aroused little interest. It was the
theorem itself, which I had thought would be casually accepted as a
straightforward codification of the DZN results, which caused surprise, and
some skepticism. ('A remarkable result, if it is right' was one comment.) It
appeared that the DZN paper was either not well known or had not found
wide acceptance among British relativists.
Only Misner had no qualms about accepting the result. t His opening
comment in the discussion period was, 'I suppose this could mean that
collapses with angular momentum are also very restricted' . I agreed that it
looked very plausible that the unique stationary fields with event horizons
were the Kerr family. But since I was wedded to my own interpretation of
this at the time, it is to M isner that the credit is due for the first apprehension
of the 'no-hair conjecture' with its correct dynamical interpretation.
In the published version of the theorem I concluded (trying to hedge my
t Like me, he was predisposed to believing it by his previous experience with
Schwarzschild-like particles in external fields (Manasse, 1963). M isner told me that he
had actually attempted to prove such a theorem in collaboration with C. V.
Vishveshwara , but they had , as he put it, 'given up too soon '.
258 W Israel
bets) :
. . . only two alternatives would be open - either the body has to divest
itself of all quadrupole and higher moments by some · mechanism
(perhaps gravitational radiation), or else an event horizon ceases to
exist.
(Israel, 1967a)
In an article in Nature, I expressed my concerns more freely :
Because of the slowing-down of processes for the external observer, it
seems reasonable to suppose . . . that it is permissible to treat the
limiting external field as static. The foregoing theorem on singular
event horizons then applies. Before reaching a singularity at the event
horizon, the star will respond to the precursory large tidal forces in such
a way as to subdue, if possible, the rise of the singularity itself. There are
two possibilities.
The first is that the star shakes off its quadrupole and higher
moments, and is, so to speak, patted into spherical shape before
crossing the event horizon. Some such idea has been suggested
previously.t The chief difficulty here is to find a plausible mechanism
. . . which �ill accomplish this task in the finite proper time available.
The second possibility, which has not been considered before, is that
the event horizon becomes smudged out and obliterated.
(Israel � 1967b)
This argument was an object lesson in how to arrive at nonsense by
stretching the frozen star picture beyond its proper limits. What is important
to remember is that it was not then clear just where these limits were, and no
overall picture was available. Perhaps the biggest conceptual hurdle then
facing the theorists was to reconcile the duality of the frozen star picture and
the local dynamical view of free fall. It was like knowing about
complementarity without having command over the mathematical
apparatus of quantum mechanics. In some ways it was worse, because in
problems like the decay of multipole moments both aspects of the duality
were involved.
The naive version of the frozen star picture was essentially this : as the star
freezes and the last gravitational waves depart,t the external field reduces to
t Here I was uncomfortably aware that DZN had not said exactly this . On the other
hand , I did not know exactly what it was they had said , much less how to express it
succinctly !
t With entrance t o the hole apparently obstructed b y the frozen star, it required some
stretch of imagination to appreciate that gravitational radiation could freely propagate
inwards as well as outwards .
its Coulomb-like part , which cannot fade away because it is anchored in the
source still lingering just outside its gravitational radius. The guiding model
is , of course, the electromagnetic field , where wavelike (transverse)
components can be self-sustaining (i.e. source-free) but Coulomb-like
(longitudinal) fields require a source. The defect of this naive version is that
in a non-linear theory like gra vita ti on a sufficiently strong field can be its
own source and thus even Coulomb-like fields can be self-sustaining. It
becomes possible to conceive that in a collapse the field and the material
source completely part company, and their multi pole moments become
independent , as DZN (read with hindsight) seem perhaps to have implied.
But a heuristic picture which needs such a sophisticated re-interpretation
becomes practically almost useless. What the frozen star picture was in the
last analysis telling us was that the frozen star is completely irrelevant to the
final state !
There would be little point to this tedious rehash of personal errors
and confusions if I did not think that it reflects, however distortedly, the
significant shift that occurred during 1967- 1969 in the general way of
thinking about the end-state of a gravitational collapse. Although the idea of
horizon instability found a few echoes in the literature (e.g. Janis et al., 1968 ;
Bel , 1969), generally the frozen star was giving way to a new image : an
elemental , self-sustaining gravitational field which has severed all causal
connection with the material sou rce that created it, and settled, like a soap
bubble,t into the simplest configuration consistent with the external
constraints. It was, above all , Roger Penrose, at least in England, who
inspired and guided the transition to this radically new viewpoint , so aptly
encoded in Wheeler's ( 1968) coincidental and timely coinage of the term
'black hole' .t
When we met in London near the end of 1967, Penrose explained this view
to me essentially as follows :
Doubts have frequently been expressed concerning (the no-hair
conjecture), since it is felt that a body would be unlikely to throw off all
its excess multipole moments just as it crossed the Schwarzschild
radius. But . . . I would certainly not expect the body itself to throw off
its multipole moments. On the other hand, the gravitational field itself
has a lot of settling to do after the body has fallen into the 'hole'. The
t This analogy is, I believe, essentially due to Misner (Thorne, 1970, 1972).
t The term was casually slipped into an address to the New York meeting of the
American Association for the Advancement of Science on 29 December 1967. Wheeler
had first used it a few months earlier in informal discussions at a conference at the
Institute for Space Studies o rganized by V. Canuto.
260 W Israel
asymptotic measurement of the multipole moments need have very

little to do with the detailed structure of the body itself; the field can
contribute very significantly. In the process of settling down, the field
radiates gravitationally and electromagnetically too, if
electromagnetic field is present. Only the mass, angular momentum
and charge need survive as ultimate independent parameters.
(Penrose, 1969)
I was fully converted to this view when , in the fall of 1969 , our group at the
University of Alberta carried out numerical integrations of idealized
collapse models with small departu res for spherical symmetry (de la Cruz et
al., 1970). Our results showed that the asymmetries are indeed radiated
away , mostly downwards through the horizon. I n effect , once the body that
originally supported the asymmetric field energies against gravity has
collapsed, the asymmetries have no alternative but to fall in after it (Israel ,
197 1). These calculations were greatly refined and extended in the work of
Richard Price ( 1972), which revealed the details of how the multi pole
moments asymptotically decay with time.
In this way a coherent picture of the terminal stage of the evolution of a
generic gravitational collapse emerged even before the no-hair conjecture on
which it was founded was fully demonstrated. The proof involved technical
complications that occupied several theorists in the next few years. My proof
was restricted to static black holes, charged or uncharged (Israel , 1967,
1 968). For stationary black holes the first key result was Carter's (197 1 )
proof that the uncharged axisymmetric ones could depend only on two
parameters (mass and angular momentum) , which was compelling evidence
that the Kerr black holes were probably the only ones. (The argument was
clinched by David C. Robinson ( 1975).) Finally, Hawking ( 1972)
demonstrated that all stationary black holes must be either static or
axisymmetric, which meant that the preceding theorems had covered all
possible cases. t
Those who began their groping exploration of the mysteries of
gravitational collapse in the early 1960s would scarcely have credited that
within a few years the essential features would be so elegantly and
completely mapped out. Nevertheless, two key problems remain. One, the
proper formulation and proof of a cosmic censorship principle, is by general
consent probably of a purely technical character and will have to await the
t Actually it was not until 1983 that the charged axisymmetric case was settled ,
independently by P . 0 . M azur and G . Bunting. Their methods also greatly simplified
and clarified the structure of the proofs.
development of more sophisticated mathematical methods. The other, the

question of what happens near the singularity predicted by the classical
Einstein theory, is far deeper :
When eventually we have a better theory of nature, then perhaps we can
try our hands, again, at understanding the extraordinary physics which
must take place at a space-time singularity.
(Penrose, 1 969)
7. 10 Towards the quantum era : the thermodynamics of black holes

(1970-74)
In those days in 1973 when I was often told that I was headed the wrong
way, I drew some comfort from Wheeler's opinion that 'black hole
thermodynamics is crazy, perhaps crazy enough to work'.
(Bekenstein, 1 980)
The inaugural conference of the newly formed European Physical Society,
held in Florence in the second week of April 1969, drew more than 800
participants who were treated to reviews by leading specialists to the current
state of the art in virtually every active branch of physics. Penrose's ( 1969)
contribution was a survey of gravitational collapse which for the first time
consolidated what had been learnt since the mid- 1960s from a coherent
viewpoint and articulated the key problems. For specialists and non
specialists alike this lecture opened new vistas : the Penrose-Hawking
singularity theorems and the no-hair conjecture were graphically described
and their significance unfolded ; the 'cosmic censor, who forbids the
appearance of naked singularities, clothing each one in an absolute event
horizon', made his debut. Of the many imaginative new insights in this
lecture not the least was the notion that the energy content of a black hole is
(in principle) not irrevocably lost if the hole has angular momentum. This
idea launched two distinct lines of development - the thermodynamics and
the quantum mechanics of black holes - which dramatically reconverged in
1974 with Stephen Hawking's discovery that black holes radiate thermally
by a quantum tunneling process.
The details of Pen rose's energy extraction process need not detain us here.
Essentially, it exploits a layer just outside the event horizon (dubbed the
'ergosphere' by Wheeler and Ruffini) where (negative) gravitational
potential energies are still sufficiently large that negative total energies are
possible for particles whose motion lags behind the inertial dragging effect of
the black hole's angular momentum . (Only spinning black holes possess an
ergosphere.) The Penrose process imagines a particle which has been
262 W Israel
injected into this region splitting into two particles . One has negative energy
and drops into the hole; the other escapes to the outside with. greater energy
than the particle that was injected. The nett effect is an extraction of some of
the hole's rotational energy, with a corresponding reduction of its mass and
angular momentum.
A 2 1-year-old Princeton graduate student , Demetrios Christodoulou
( 1970), undertook a detailed study of the efficiency of quasistatic processes of
this type, in which the black hole's parameters are changed gradually by
injecting a succession of test particles. He found that the efficiency is limited
by an inequality which, in effect, states that a certain quantity can never
decrease. This is a function of the mass and angular momentum of the black
hole which Christodoulou called the 'irreducible mass'; its simple
geometrical significance was not yet apparent.
At the Texas Symposium in Austin in December 1 970 , Stephen Hawking
( 197 1) crowned the edifice of classical black hole theory with his elegant
demonstration that Christodoulou's result was a special case of a
very general law. Hawking's theorem states that the area of the event
horizon can never decrease in any interaction of a black hole with its
environment, provided the energy density of any accreted material is non
negative.
The obvious analogies with thermodynamics al ready foretokened in the
title of Christodoulou's ( 1970) paper, 'Reversible and i rreversible
transformations in black-hole physics' , were extensively developed by
Bardeen , Carter and Hawking during the Les Houches Summer School on
black holes in August 1972 (Carter, 1973). Black hole analogues were
formulated for all four of the laws of thermodynamics, with the black hole's
area playing the role of entropy and its surface gravity the role of
temperature.
However, there was almost universal agreement that this analogy was
purely formal :
In fact the effective temperature of a black hole is absolute zero . One
way of seeing this is to note that a black hole cannot be in equilibrium
with black body radiation at any non-zero temperature, because no
radiation could be emitted from the hole whereas some radiation would
always cross the horizon into the black hole.
(Bardeen, Carter and Hawking , 1973)
The real effective temperature of a black hole is well defined and
unambiguously zero, as also are its chemical potentials. The ordinary
particle conservation laws, and the ordinary second law of
thermodynamics are unquestionably transcended by a black hole, in

the sense that particles and entropy can be lost without trace from an
external point of view. It is not possible to mitigate this transcendence
by somehow relating the amount of entropy (or the number of particles)
which have gone in to the subsequent increase in surface area.
(Carter, 1973)
These admonitions were partly intended for one person in particular,
a young Les Houches student from Princeton, Jacob D. Bekenstein ( 1972 ,
1973), who was advocating the heretical idea that the area of a black hole
actually was the same thing as its entropy , apart from a universal
proportionality factor. t This proposal had no takers and was up against a
seemingly impregnable counterexample given by Robert Geroch at a
colloquium in Princeton in December 197 1 , which showed that the entropy
of a black hole can be increased arbitrarily without changing its area : Lower
a weightless box containing radiation to the horizon of a spherical black
hole, extracting all of its mass-energy as work. Now open the box and allow
the radiation to drain into the hole : the entropy of the hole grows, but its
mass (and therefore surface area) remains unchanged . t
With the benefit o f hindsight it i s easy to pinpoint the underlying difficulty
with Bekenstein's proposal . As he conceived it , it was not quite 'crazy'
enough. For all his radicalism concerning the interpretation of black hole
area, he hesitated to relinquish the conventional view that the surface
gravity had nothing to do with the real temperature of the black hole :
But we emphasize that one should not regard [the surface gravity] as
the temperature of the black hole : such an identification can easily lead
to all sorts of paradoxes, and is thus not useful.
(Bekenstein, 1973)
The struggle to keep this inherently contradictory scheme afloat led him
into convoluted arguments which, at best , were just short of compelling and,
in less fortunate instances, an easy target for critics (e.g. Israel , 1973).
Help was to come from an unexpected quarter, again traceable to
Penrose's energy-extraction process. Zel'dovich ( 1970) and also Misner
( 1972)t pointed out that there is a w ave analogue of this process, called
.
t Bekenstein's ( 1980) personal reminiscences offer a vivid account of this period.
t This paradox was not cleared up until ten years later, when W. G. Unru h and
R. M . Wald ( 1 982) showed that not all of the mass-energy can be extracted as work
because of buoyancy effects due to the bath of Hawking radiation enveloping the hole.
t At this time Misner was examining various possibilities for an amplification mechanism
involving a massive black hole at the galactic centre in an effort to explain the
gravitational wave observations of his colleague at Maryland , Joseph Weber.
264 W Israel
'superradiant scattering', in which an impinging wave mode is amplified as it

scatters off a rotating black hole if its energy per unit angular momentum is
smaller than the angular velocity of the hole. Regarding this as a form of
induced emission, Zel'dovich argued that there should be spontaneous
emission in the same modes by virtue of the well-known Einstein relations of
the quantum theory of radiation. In other words, a spinning black hole
"hould spontaneously emit radiation from its ergosphere and, in the process,
slow down until it comes to a stop and the ergosphere disappears ; then ,
according to Zel'dovich , the emission should stop. Quantum field
theoretical calculations by Zel' dovich and Starobinsky, using a somewhat
ad hoc renormalization prescription , supported this prediction.
Here is Stephen Hawking's (1987) account of what happened next :
If black holes have entropy, then they also ought to have a temperature
. . . so black holes ought to emit radiation. But by their very definition,
black holes are objects which do not emit anything. It therefore seemed
that the identification of the area of the event horizon of a black hole
with entropy could not be correct. In fact, in 1972 , I wrote a paper with
Brandon Carter and an American colleague, Jim Bardeen , in which we
pointed ou_t that , although there were many similarities between
entropy and the area of the event horizon , there was this apparently
fatal difficulty. I must admit that in writing this paper I was motivated
partly by irritation with Bekenstein who, I felt, had misused my result
about the increase of the area of the event horizon. However, in the end
it turned out that he was basically correct.
In September 1973 I visited Moscow.t While I was there I discussed
black holes with two of the leading Soviet experts. Yakov Zel' dovich
and Alexander Starobinsky. They convinced me that the quantum
mechanical Uncertainty Principle implied that rotating black holes
should create and emit particles. I believed their arguments on physical
grounds but I did not like the mathematical way in which they
calculated the emission . I therefore set about devising a better
treatment of the emission . I described the formulation of this treatment
at an informal seminar that I gave at Oxford at the end of November
1973. At that time I had not done the calculations to find out how much
would actually be emitted . I was expecting to discover just the radiation
that Zel'dovich and Starobinsky had predicted from rotating black
t This was the first meeting of Hawking and Zel'dovich. When they were introduced,
Hawking is reputed to have said , 'You are Zel'dovich? I thought you were a
corporation'. This story is, unfortunately, apocryphal .
Da rk sta r s: the evolution of an idea 26�
holes. However, when I did the calculation , I found, to my surprise and

annoyance, that even non-rotating black holes should apparently
create and emit particles at a steady rate. At first I thought that this
emission was a pathological result which indicated that �me of the
approximations that I had used in my treatment was not valid . I was
afraid that if Bekenstein found out about it, he would use it as a fu rther
argument to support his ideas about the entropy of black holes, which I
still did not like . However, the more I thought about it, the more it
seemed that the approximations ought to hold . What finally convinced
me that the emission was a real effect was that the spectrum of the
emitted particles was exactly that which would be emitted by a hot
body , and that the black hole was emitting particles at just the rate
required to prevent violations of the Second Law . Since then the
calculations have been repeated in a number of different forms by other
people. They all confirm that black holes ought to emit particles and
radiation as if they were hot bodies with a temperature which is higher
the smaller the mass of the black hole.
Hawking's preliminary report, 'Black hole explosions?', submitted to
Nature on January 1 7 , 1974, aroused strong opposition almost as soon as it
was in print (Davies and Taylor, 1974) ; skepticism was prolonged and
virtually unanimous . . Over the next two years, as various alternative
derivations appeared (e .g. Hartle and Hawking, 1976) which were less
unconventional from the quantum field theorist's point of view, it slowly
became clear that Hawking's result was genuine and inescapable.
The discovery that , at the quantum level , black holes are
thermodynamical black bodies at one stroke crowned a classical theory
which had reached an undreamed-of level of elegance and sophistication
and opened the door to new horizons in the quantum domain . Many
perceived in this mysterious result a clue to some possible future synthesis of
general relativity , quantum theory and thermodynamics whose nature we
cannot yet comprehend .
But science is not just a catalogue of ascertained facts about the
universe ; it is a mode of progress, sometimes tortuous, sometimes
uncertain . And our interest in science is not merely a desire to hear the
latest facts added to the collection ; we like to discuss our hopes and
fears, probabilities and expectations. I have told the detective story as
far as it has unrolled itself. I do not know whether we have reached the
last chapter.
(Eddington , 1927)
266 W Israel
References
Adams, J. B . , Harrison, B. K., Klauder, L. T. Jnr, Mjolsness, R., Wakano, M . , Wheeler,

J. A. and Willey, R. ( 1958). Some implications of general relativity for the structure
and evolution of the universe. La Structure et £Evolution de L. Univers, Onzieme
Conseil de Physique, Institut International de Physique Solvay. R . Stoops, Brussels,
pp. 97- 146.
Adams, W. S. ( 19 1 5). The spectrum of the companion of Sirius. Pub/. Astron. Soc.
Pacific, 27, 236-7.
Adams, W. S . ( 1925). The relativity displacement of the spectral lines in the comparison
of Sirius. Proc. Nat. Acad. Sci. , 1 1 , 382-7.
Ambartsumian, V. A. ( 1958). On the evolution of galaxies. Contribution to 1 1th Solvay
Congress, pp. 24 1-79 (see Adams et al. , 1958).
Anderson, A. ( 1920). On the advance of the perihelion of a planet, and the path of a ray
of light in the gravitational field of the Sun. Phil. Mag . , 39, 626-8 .
Anderson , J. L. and Gautreau , R . ( 1966) . Possible causal violations at radii greater than
the Schwarzschild radius. Phys. Lett. , 20, 24-5.
Anderson , W. ( 1 929). Uber die Grenzdichte der Materie und der Energie . Z. Phys., 56,
8 5 1--6.
Anderson , W. ( 1936) . Existiert eine obere Grenze fiir die Dichte der Materie und der
Energie? Acta Comment. Univ. Tartuensis, A29, 1- 142.
Baade, W. ( 1952) . Letter to Bernard Lovell (26 May) . Reprinted in Smith and Lovell
( 1983).
Baade, W. and M inkowski, R . ( 1954). On the identification of radio sources. Astrophys.
J., 1 19, 2 1 5-3 1 .
Baade, W . and Zwicky, F . ( 1934) . O n supernovae; Cosmic rays from supernovae. Proc .
Nat. Acad. Sci., 20, 254--63.
Bardeen, J. M . , Carter, B . and Hawking, S . W. ( 1973). The fou r laws of black hole
mechanics. Commun. Math . Phys., 31 , 1 6 1-70.
Baym, G. and Pethick, C. ( 1979). Physics of neutron stars. Ann. Rev. Astron. Astrophys. ,
1 7 , 4 1 5-43.
Beals, C. S. ( 1929). On the nature of Wolf-Rayet emission . Mon. Not. R. Astron. Soc.,
90, 202- 12.
Beckedorff, D . L . ( 1962). Terminal configurations of stellar evolution . A.B. Senior thesis,
Princeton [reported in Harrison et al. ( 1965), p . 126] .
Begelman, C., Blandford , R . D . and Rees, M . J . ( 1 980). Massive black hole binaries in
active galactic nuclei . Nature, 287 , 307-9.
Bekenstein, J. D . (1972) . Black holes and the second law. Lett. Nuovo Cim., 4 , 737-40.
Bekenstein , J. D . ( 1973). Black holes and entropy . Phys. Rev., D7, 2333-46.
Bekenstein, J. D. ( 1980) . Black hole thermodynamics. Physics Today (January) ,
pp. 24-3 1 .
Bel , L . ( 1 969) . Schwarzschild singularity. J . Math. Phys ., 1 0 , 150 1-3 .
Belinfante, F. J . ( 1 966) . Kruskal space with wormholes. Phys. Lett., 20, 2 5-7 .
Bell Burnell, S . Jocelyn ( 1977). Petit Four (8th Texas Symposium on Relativistic
Astrophysics). Ann. NY Acad. Sci., 302, 685-9.
Bergmann, P . G . ( 1942). An Introduction to the Theory of Relativity, pp. 203-4.
Prentice-Hall : New York.
Bethe, H. A. ( 1939). On energy generation in stars. Phys. Rev., 55, 1 03(L), 434-56.
Biermann, L. ( 1932) . Untersuchungen iiber den inneren Aufbau der Sterne.

IV. Konvektionszonen im Innern der Sterne . Z. Astrophys . , 5, 1 1 7-39.
Bolton, C. T. ( 1972). Identification of Cygnus X- 1 with HOE 226868 . Nature, 235,
27 1-3 .
Boyer, R . H . and Lindquist, R . W . ( 1967). Maximal analytic extension of the Kerr
metric . J. Math . Phys . , 8, 265-8 1 .
Brown , R . Hanbury ( 196 1). Cited i n Hoyle ( 1 98 1), p . 1 3 .
Burbidge , G. R . ( 1956). On synchrotron radiation from Messier 87. Astrophys. J., 124 ,
4 16-29.
Burbidge, G. R . ( 1958). Possible sou rces of radio emission in clusters of galaxies.
Astrophys. J., 128 , 1-8 .
Burbidge, G. R . ( 1963) . The detection of stars with neutron cores. Astrophys. J., 137,
995-6.
Cameron , A . G. W . ( 1 959). Neutron star models. Astrophys. J., 130, 884--94.
Campbell, W . W . ( 19 1 3). Stellar Motions, p. 242 . Yale University Press : New Haven.
Carter, B. ( 1966) . Complete analytic extension of the symmetry axis of Kerr's solution of
Einstein's equations. Phys. Rev., 141 , 1242-7 .
Carter, B . ( 1968) . Global structure of the Kerr family of gravitational fields. Phys. Rev.,
1 74 , 1559-7 1 .
Carter, B. ( 197 1 ) . A n axisymmetric black hole has only two degrees o f freedom. Phys.
Rev. Lett., 26, 33 1-3 .
Carter, B . ( 1973). Black hole equilibrium states. In DeWitt ( 1973) , pp. 56--2 14.
Chandrasekhar, S . ( 193 la). The maximum mass of ideal white dwarfs. Astrophys. J., 74,
8 1-2 .
Chandrasekhar, S . ( 193 1b). The highly collapsed configurations of a stellar mass. Mon.
Not. R. Astron . Soc. , 9 1 , 456-66 .
Chandrasekhar, S . ( 1932) . Some remarks on the state of matter in the interior of stars.
Z. Astrophys., 5, 32 1-7 .
Chandrasekhar, S. ( 1934a). On the hypothesis of the radial ejection of high-speed atoms
from the Wolf-Rayet stars, and the novae. Mon. Not. R . Astron. Soc., 94, 522-3 8 .
Chandrasekhar, S . ( 1934b) . Stellar configurations with degenerate cores. Observatory, 57,
373-77.
Chandrasekhar, S . ( 1935a). The highly collapsed configurations of a stellar mass (second
paper) . Mon. Not. R . Astron. Soc., 95, 207-25 .
Chandrasekhar, S . ( 193 5b) . Stellar configurations with degenerate cores (second paper).
Mon. Not. R. Astron. Soc., 95, 676--9 3.
Chandrasekhar, S . ( 1 9 5 1 ) . The structure, the composition, and the source of energy of
the stars. In Astrophysics, ed . J . A . Hynek. McGraw-Hill : New York.
Chandrasekhar, S . ( 1 969). Some historical notes. Am. J. Phys. , 37, 577-84.
Chandrasekhar, S. ( 1972) . The increasing role of general relativity in astronomy (Halley
Lecture). Observatory, 92, 160-74.
Chandrasekhar, S. ( 1977). Interview with Spencer Weart (Niels Bohr Library, American
Institute of Physics), cited in Greenstein ( 1984), Chapter 12.
Chandrasekhar, S. ( 1980). The role of general relativity in astronomy : retrospect and
prospect. In Highlights of Astronomy, vol . 5, ed . P . A . Wayman, pp. 45-6 1 . Reidel :
Dord recht.
Chandrasekhar, S. ( 198 3). Eddington. The Most Distinguished Astrophysicist of his Time.
Christodoulou, D . ( 1970) . Reversible and irreversible transformations in black-hole
physics. Phys. Rev. Lett., 25, 15 9 6--7 .
268 W Israel
Clerke, Agnes M. ( 1 903). Problems in Astrophysics, p. 400 . A. and C. Black : London .

Cooperstock, F. I. and Junevicus, G. J . G. ( 1 973). Perspectives on the Schwarzschild
singularity. Nuovo Cimento, 168, 387-97 .
·
Cowley, A . , Crampton, D . , Hutchings, J . B . , Remillard , R . and Penfold , J . ( 1983).

Discovery of a massive unseen star in LMC X-3 . Astrophys. J., 272, 1 18-22.
Cowling, T. G. ( 1934). The stability of gaseous stars. Mon. Not. R. Astron. Soc. , 94,
768-82.
Cruz, V. de la, Chase, J. E. and Israel , W. ( 1 970). Gravitational collapse with
asymmet ries . Phys. Rev. Lett., 24, 423-6.
Cruz, V . de la and Israel, W . ( 1967). Gravitational bounce. Nuovo Cimento, 5 1 A ,
744-60.
Curtis, A. R . ( 1950). The velocity of sound in general relativity , with a discussion of the
problem of the fluid sphere with constant velocity of sound . Proc. R. Soc., A200,
248--6 1 .
Dantzig, D . van ( 1939). Stress tensor and particle density in special relativity theory.
Nature, 143, 855-6.
Davies, P. C. W . and Taylor, J. G. ( 1974). Do black holes really explode? Nature, 250,
37-8 .
Descartes, R . ( 1644). Principiorum Philosophae, Part III, Chapter 52 .
Detweiler, S . ( 1 982). Black Holes: Selected Reprints. Am . Assoc. of Physics Teachers,
Stony Brook , NY.
Dirac, P . A. M., Peierls, R . and Pryce, M . H. L . (1942). On Lorentz invariance in the
quantum theory. Proc. Camb. Phil. Soc., 38, 193-200.
Doroshkevich , A. G . , Zel'dovich, Ya. B. and Novikov, I . D. ( 1965). Gravitational
collapse of nonsymmetric and rotating masses. Sov. Phys. JETP, 22, 122-30.
Douglas, A. Vibert ( 1956). The Life of Arthur Stanley Eddington. Thomas Nelson :
London.
Droste, J. ( 19 16). The field of a single centre in Einstein's theory of gravitation , and the
motion of a particle in that field . Proc. Acad. Sci. (Amsterdam), 19, 197-2 15.
Eddington , A . S. ( 19 18). On the conditions in the interior of a star. Astrophys. J., 48,
205- 13.
Eddington, A. S. ( 1 920). Space, Time and Gravitation, p . 98. Cambridge U niversity Press :
Cambridge.
Eddington, A. S . ( 1 92 1). Das Strahlungsgleichgewicht der Sterne. Z. Phys . , 7 , 35 1-97.
Eddington, A . S. ( 1924a). On the relation between the masses and luminosities of the
stars. Observatory, 47, 107-14.
Eddington, A . S. (1924b, 1925). Correspondence with W. S . Adams, cited in Douglas
( 1956), pp. 75-7.
Eddington, A. S. ( 1924c). A comparison of Whitehead's and Einstein's formulas. Nature,
1 13, 192.
Eddington , A. S . ( 1926). The Internal Constitution of the Stars, p. 1 72. Cambridge
Eddington, A. S. ( 1927). Stars and Atoms, pp. 37, 50. Cambridge University Press :
Cambridge.
Eddington, A. S. ( 1 929). RAS Meeting , Nov . 8. Observatory, 52, 345-54.
Eddington, A. S. (1935a). Relativistic degeneracy . Observatory, 58, 37-9.
Eddington , A. S. ( 1935b). On 'relativistic degeneracy'. Mon. Not. R. Astron. Soc., 95,
194-206.
Eddington , A. S . ( 1936a). Relativity Theory of Protons and Electrons, §§ 13.3, 13.5, 1 3 .8 .
Eddington, A. S. ( 1936b) . Lecture , Harvard Tercentenary Conference, quoted in

Chandrasekhar ( 1 983), p. 47.
Eddington , A. S. ( 1939). The hydrogen content of white dwarf stars in relation to stellar
evolution . Mon . Not. R. Astron . Soc . , 99, 595-606.
Eddington , A. S . ( 1940). The physics of white dwarf matter. Mon. Not. R . Astron. Soc .,
100 , 582-94.
Einstein , A. ( 1939). On a stationary system with spherical symmetry consisting of many
gravitating masses. Ann . Math . (Princeton), 40, 922-36.
Einstein , A . and Rosen, N . ( 1935). The particle problem in general theory o f relativity.
Phys. Rev . , 48, 73-7.
Eisenstaedt, J . ( 1982) . His to ire et singularites de la solution de Schwarzschild
( 19 1 5- 1923), Arch. Hist. Exact Sci. , 27 , 1 57-98 .
Faulkner, J . , Hoyle, F. and Narlikar, J . V . ( 1 964) . On the behaviour of radiation near
massive bodies. Astrophys . J., 140, 1 100-5.
Finkelstein , D. ( 1958). Past-future asymmetry of the gravitational field of a point-
particle. Phys. Rev . , 1 10, 965-7 .
Flamm , L . ( 1 9 16). Beitrage zur Einsteinschen Gravitations theorie. Phys. Z., 17, 448-54.
Fowler, R . H . ( 1926). On dense matter. Mon . Not. R. Astron. Soc., 87, 1 14-22.
Fowler, R. H . ( 1929). Statistical Mechanics, p . 522. Cambridge University Press :
Cambridge.
Fronsdal , C. ( 1959). Completion and embedding of the Schwarzschild solution . Phys.
Rev . , 1 16 , 778-8 1 .
Gamow , G . ( 1937). Structure of Atomic Nuclei and Nuclear Transformations, pp. 234-8 .
Clarendon Press : Oxford .
Gamow , G. ( 1940) . The Birth and Death of the Sun , pp . 1 70, 189-93 . Viking Press : New
York.
Gamow, G. and Schonberg, M . ( 194 1). Neutrino theory of stellar collapse. Phys. Rev.,
5 9 , 539-47 .
Gamow, G . ( 1944) . The evolution of contracting stars. Phys. Rev., 65, 20-32.
Gaskell , C. M. ( 1985). Galactic mergers, starburst galaxies, quasar activity and massive
binary black holes . Nature, 315, 386.
Gerasomovic, B . P. ( 1 93 1). The nuclei of planetary nebulae. Observatory, 54 , 108- 10.
Geroch, R. P. ( 1970). The domain of dependence. J. Math. Phys. , 1 1 , 437-49.
Gertsenshtein , M . E. ( 1967) . The possibility of an oscillatory nature of gravitational
collapse. Sou. Phys. JETP, 24, 87-90; The nature of the central body in the
Schwarzschild solution . Ibid., 24, 754-9.
Gibbons, G . W. ( 1986). The elliptic interpretation of black holes and quantum
mechanics. Nucl. Phys., B271 , 497-508 .
Ginzburg , V. L . ( 196 1). The nature of the radio galaxies. Sou. Astron. , 5, 282-3 .
Ginzburg , V. L . ( 1964) . On the magnetic fields of collapsing masses and on the nature of
superstars . In Robinson et al. ( 1965) , pp . 283-7 ; Sou. Phys. Dokl. 9, 329-32 .
Ginzburg , V. L . and Ozemoy, L . M . ( 1965) . On gravitational collapse of magnetic stars.
Sou. Phys. JETP, 20, 689-96.
Ginzburg , V. L. and Ozemoy, L. M . ( 1977) . On the nature of quasars and active
galactic nuclei. Astrophys. Space Sci. , 48, 40 1-20 ; 50, 23-4 1 .
Ginzburg , V . L . ( 1984) . Remarks o n m y work i n radio astronomy. In Sullivan ( 1984) ,
pp. 289-302 .
Gold, T. ( 195 1). M eeting of Royal Astronomical Society, 12 Oct . 195 1 . Observatory, 72,
2 10- 14.
270 W Israel
Gold, T. (1968). Rotating neutron stars as the origin of the pulsating radio sources .
Nature, 218, 73 1-2.
Greenstein, G. ( 1984). Frozen Star. Macdonald : London.
Greenstein, J. L. ( 1963). Quasi-stellar radio sources. Scientific American, 209 (December),
54-62 .
Greenstein, J. L. ( 1 984). Optical and radio astronomers in the early years. In Sullivan
( 1984), pp. 67-8 1 .
Gullstrand, A. ( 1922). Allegemeine Losung des statischen Einkorper-problems i n der
Einsteinschen Gravitations theorie. Arkiv. Mat. Astron . Fys., 16(8), 1- 15.
Harrison, B. K., Thorne, K . S . , Wakano, M . and Wheeler, J. A . ( 1965). Gravitation
Theory and Gravitational Collapse. University of Chicago Press : Chicago.
Harrison, E. R. ( 1972). Particle barriers in cosmology. Comment Astrophys . Space Phys.,
4 , 187-92.
Hartle, J. B . and Hawking, S. W. (1976). Path-integral derivation of black-hole radiance.
Phys. Rev., 0 13, 2 1 88-203.
Hawking , S . W . ( 197 1). Gravitational radiation from colliding black holes. Phys. Rev.
Lett., 26, 1344--6 .
Hawking, S . W. ( 1972) . Black holes in general relativity. Commun . Math. Phys., 25,
152--66.
Hawking , S. W. ( 1974). Black hole explosions? Nature, 248, 30-- 1 .
Hawking, S . W . ( 1987). From the Big Bang to Black Holes: A Short History of Time.
Bantam Books : New York (in press).
Hawking, S. W. and Ellis, G. F. R. ( 1973). The Large Scale Structure of Space- Time.
Hawking, S. W. and Penrose, R. ( 1969). The singularities of gr<l;vitational collapse and
cosmology. Proc. R. Soc. (London), A314, 529-48 .
Hazard , C. ( 1985). The coming of age of QSOs. In Active Galactic Nuclei, ed . J . E.
Dyson , pp. 1- 19. Manchester University Press : Manchester.
Hazard, C . , Mackey, M . B. and Shimmins, A. J. ( 1963). Investigation of the radio source
3C273 by the method of lunar occultations. Nature, 197, 1037-9.
Hertzprung, E. ( 1 905). Zur Strahlung der Sterne (giants and dwarfs). Z. Wiss. Photog ., 3.
Engl . transl. in Source Book in Astronomy, ed. H . Shapley . Harvard University Press :
Cambridge, Mass.
Hertzprung, E. ( 19 15). Effective wavelengths of absolutely faint stars. Astrophys. J., 42,
1 1 1-19.
Hewish , A . (1975). Pulsars and high-density physics. Rev. Mod. Phys., 47, 567-72.
Hewish, A . , Bell, S. J., Pilkington , J. D . H ., Scott , P. F. and Collins, R . A . ( 1968).
Observation of a rapidly pulsating radio source. Nature, 217, 709-13.
Hey, J. S . ( 1973). The Evolution of Radio Astronomy. Neale Watson : New York .
t'Hooft, G. ( 1985). On the quantum structure of a black hole. Nucl. Phys ., B256, 727-45.
Hoyle, F. ( 1946) . The synthesis o f t h e elements from hydrogen . Mon . Not. R. Astron.
Soc., 1 06, 343-83 .
Hoyle, F. ( 1955). Frontiers of Astronomy, p . 204. Heinemann : London.
Hoyle, F. ( 198 1). The Quasar Controversy Resolved. University College Cardiff Press :
Cardiff.
Hoyle, F. and Fowler, W. A. ( 1 963). Nature of strong radio sources. Nature, 197, 533-5 .
Hoyle, F . , Fowler, W. A . , Burbidge, G . R . and Burbidge, E. M . ( 1 964). On relativistic
astrophysics. Astrophys. J., 139, 909-28 .
Hoyle, F. and Narlikar, J. V . ( 1964). On the avoidance of singularities in C-field
cosmology. Proc. R. Soc. , A278 , 465-78 .
Hund , F. ( 1936). Materie unter sehr hohen Drucken und Temperaturen . Ergebnisse
Exakten Naturwiss., 15, 189-228 .
Hund, F . ( 1 97 1). I rrwege und Hemmungen beim Werden der Quantentheorie. In
Quanten und Felder, ed . H . P . Diirr, pp. 1- 10. F. Vieweg : Braunschweig.
Israel , W . ( 1966) . Is gravitational collapse really irreversible? Nature, 2 1 1 , 466-7.
Israel , W. ( 1967a). Event horizons in static vacuum space-times . Phys. Rev., 164, 1776-9 .
Israel , W . ( 1967 b) . Possible instability in the self-closure phenomenon in gravitational
collapse. Nature, 2 16, 148-9 and 3 12.
Israel, W. ( 1968). Event ho rizons in static electrovac space-times. Commun . Math . Phys.,
8 , 245-60.
Israel, W . ( 197 1). Event horizons and gravitational collapse. J. Gen. Rel. Grav., 2, 53-6 1 .
Israel , W . ( 1973). Entropy and black hole dynamics . Lett. Nuovo Cimento, 6 , 267-9.
Israel , W. ( 1983). Black holes . Sci. Prog. (Oxford), 68, 333-63 .
Israel , W. and Khan , K . A. ( 1964) . Collinear particles and Bondi dipoles i n general
relativity. Nuovo Cimento , 33, 33 1-44.
Jaki, S. L. ( 1978). Johann Georg von Soldner and the gravitational binding of light, with
an English translation of his essay on it published in 180 1 . Found. Phys., 8, 927-50.
Janis, A. I . , Newman , E. T. and Winicour, J. ( 1968) . Reality of the Schwarzschild
singularity . Phys. Rev. Lett., 20, 878-80.
Jeans, J. H . ( 1926). On liquid stars and the liberation of stellar energy . Mon. Not. R.
Astron. Soc. , 8 7 , 400- 14.
Jeans , J. H . ( 1928). Astronomy and Cosmogony, pp. 72 , 352 and Chap . 5. Cambridge
Jeffreys, H . ( 19 18). The compressibility of dwarf stars and planets. Mon. Not. R. Astron.
Soc ., 78 , 183-4.
Jeffreys, H . ( 1 930) . Convections in stars. Mon. Not. R . Astron. Soc. , 91 , 12 1-2 .
Jiittner, F . ( 1 928) . Die relativistische Quantentheorie des idealen Gases. Z. Phys., 47,
542-66.
Kardashev, N . S. ( 1964) . M agnetic collapse and the nature of intense sources of cosmic
radio-frequency emission . Sov. Astron. AJ, 8, 643-8 .
Kerr, R . P . ( 1 963). Gravitational collapse and rotation . In Robinson , Schild and
Schucking ( 1965), pp. 99- 102.
Khalatnikov, I . M . and Lifshitz, E. M . ( 1970). General cosmological solution o f the
gravitational equations with a singularity in time. Phys. Rev. Lett. , 24, 76-9.
Kiepenheuer, K . 0. ( 1950). Cosmic rays as the source of general galactic radio emission.
Phys. Rev., 79, 738.
Komar, A . ( 1965). Bootstrap gravitational geons. Phys. Rev. , 137B, 462-6.
Kosirev, N . A. ( 1934). Radiative equilibrium of the extended photosphere. Mon. Not. R.
Astron. Soc., 94, 430-43.
Kretschman, E. ( 1923). Das statische Einkorperproblem in der Einsteinschen Theorie.
Antwort an Hrn A. Gullstrand . Arkiv Mat. Astron. Fys., 17, 1-4.
Kruskal, M . D . ( 1960). M aximal extension of Schwarzschild metric. Phys. Rev., 1 19,
1743-5.
Landau, L. ( 1932). On the theory of stars. Phys. Z. Sowjetunion, 1 , 285-7 .
Landau , L. ( 1938). The origin of stellar energy . Nature, 141 , 333-4.
Laplace, P. S. ( 1796) . Exposition du Systeme du Monde, vol . 2, p. 305 . J. B. M . Duprat :
Paris. See also Hawking and Ellis ( 1973), pp. 365-8 .
Laue, M . von ( 1953). Die Relativistats theorie (3rd ed .), vol . 2, p . 135. F. Vieweg :
Braunschweig .
Lemaitre, G . ( 1933). L 'univers en expansion. Ann. Soc. Sci. (Bruxelles), A53, 5 1-8 5.
272 W Israel
Liebert, J. ( 1980). White dwarf stars. Ann. Rev. Astron . Astrophys. , 18, 363-98 .
Lifshitz, E. M . , Sudakov, V . V. and Khalatnikov, I . M . ( 196 1). Singularities of
cosmological solutions of gravitational equations III. Sov. Phys. J ET.P, 13, 1298-303.
Lifshitz, E. M. and Khalatnikov, I . M. ( 1963). Problems of relativistic cosmology. Sov.
Phys. Uspekhi, 6, 495-522.
Lindquist, R. W. and Wheeler, J. A. ( 1957) . Dynamics of a lattice universe by the
Schwarzschild-cell method . Rev. Mod. Phys., 29, 432-43 .
Lodge, Sir Oliver ( 1920). Discussion on theory of relativity. Mon. Not. R . Astron. Soc.,
80, 96- 1 1 8 .
Lodge, S i r Oliver ( 192 1). On the supposed weight and ultimate fate o f radiation. Phil.
Mag., 41 , 549-57 .
Lucretius (55 BC). On the Nature of the Universe, p . 59. Penguin Books :
Harmondsworth.
Lynden-Bell, D. ( 1969). Galactic nuclei as collapsed old quasars. Nature, 223, 690-4.
McCormmach, R. ( 1968). John M ichell and Henry Cavendish : weighing the stars. Brit .
J. Hist. Sci., 4 , 126-55.
McCrea, W. H. ( 1964). Release of gravitational energy in general relativity. Nature, 201 ,
589.
Manasse, F. K. ( 1963). Distortion in the metric of a small centre of gravitational
attraction due to its proximity to a very large mass. J. Math. Phys. , 4, 746-6 1 .
Matthews, T . A . and Sandage, A . R . ( 1 963). Optical identification of 3C48, 3C 1 96 and
3C286 with stellar objects. Astrophys. J., 138 , 30-56.
Menzel, D . H . ( 1926). The planetary nebulae. Pub/. Astron. Soc. Pacific, 38, 295--3 12 .
Michell , John ( 1784). O n the means of discovering the distance, magnitude, etc . , o f the
fixed stars, in consequence of the diminution of their light, in case such a diminution
should be found to take place in any of them, and such other data should be procured
from observations, as would be further necessary for that purpose. Phil. Trans. R . Soc.
(London), 74, 35-57 . Reprinted in Detweiler ( 1982) .
Milne, E. A . ( 1930). The analysis of stellar structure. Mon. Not. R. Astron. Soc. , 9 1 ,
4-55 ; Observatory, 53, 305-8 .
Milne, E. A. ( 19 3 1 ) . On dense stars. Observatory, 54, 140-5.
Minkowski, R. ( 1942). The Crab nebula. Astrophys. J., 96, 199-2 1 3 .
Minkowski, R. ( 196 1 ) . Letter. Sky and Telescope, 21 , 2 3 .
Misner, C. W . ( 1967). Taub-Nut space a s a counterexample t o almost anything. I n
Relativity Theory and Astrophysics, vol 1 , ed . J . Ehlers, pp. 160-7 . American
Mathematical Society : Providence, Rhode I sland.
Misner, C. W. and Wheeler, J. A. ( 1957). Classical physics as geometry : gravitation,
electromagnetism, unquantized charge and mass as properties of empty space . Ann.
Phys., 2, 525-603.
Misner, C. W. ( 1969). M ixmaster universe. Phys. Rev. Lett. , 22, 107 1-4.
Misner, C. W. ( 1972). Interpretation of gravitational wave observations. Phys. Rev. Lett.,
28, 994-7.
M0ller, Chr. and Chandrasekhar, S. ( 193 5). Relativistic degeneracy. Mon . Not. R .
Astron. Soc. , 95, 673-6.
Morrison , P . ( 1965). Summary. In Robinson, Schild and Schucking ( 1965), pp. 437-8 .
Morrison, P . ( 1977). Astronomy and the laws of physics . In Highlights of Astronomy,
vol . 4, part 1 , ed . E. A. Muller, p. 35. Reidel : Dordrecht.
Mysak, L . A. and Szekeres, G. ( 1966). Behaviour of the Schwarzschild singularity in
superimposed gravitational fields. Can. J. Phys., 44 , 6 1 7-27 .
Ne'eman, Y . ( 1965). Expansion as an energy source in quasi-stellar radio sources.

Astrophys. J., 141 , 1303-5.
Newcomb, S. ( 1 908). The Stars: A Study of the Universe, p. 205. John Murray : London .
Novikov. I . D. ( 1964). Delayed explosion of a part of the Friedman universe and
quasars . Sov. Astron. AJ, 8, 857-63.
Novikov, I. D. ( 1966). Change of relativistic collapse into anticollapse and kinematics of
a charged sphere. Sov. Phys. JETP Lett., 3, 142-4.
Observatory, The ( 19 16). Notes : The spectrum of the companion of Sirius. 39 , 142-3 .
Observatory, The ( 1 9 1 7). Proceedings at meeting of the Royal Astronomical Society,
Friday, 8 Dec. 19 16. 40 , 3 5-49.
O'Dell . C. R. ( 1963). The evolution of the central stars of planetary nebulae. Astrophys.
J., 1 38, 67-78.
O pik , E. J . ( 1 9 16). On the densities of visual binary stars. Astrophys. J . 44, 292-302.
.
Oppenheimer, J. R. and Serber, R. ( 1938). On the stability of stellar neutron cores. Phys.
Rev . , 54, 540.
Oppenheimer, J. R. and Snyder, H . ( 1 939). On continued gravitational contraction.
Phys. Rev . , 56, 455-9.
Oppenheimer, J . R. and Volkoff, G. ( 1939). On massive neu tron cores. Phys. Rev., 55,
374-8 1.
Pacini , F . (1 967). Energy emission from a neu tron star. Nature, 216, 567-8 .
Painleve, P . (192 1). La M ecanique classique et la theorie de la relativite. C.R. Acad. Sci.
(Paris), 173, 677-80.
Pais, A. ( 1982) . Subtle is the Lord . . . The Science and the Life of Albert Einstein, p. 509.
Clarendon Press : Oxford.
Papapetrou, A. ( 1966). Champs gravitationnels stationnaires a symetrie axiale. Ann.
Inst. H . Poincare, A IV, 83- 105.
Payne-Gaposchkin, C. (1957). The Galactic Novae, p . 3 12 . North-Holland : Amsterdam.
Peierls, R. (1936). Note on the derivation of the equation of state for a degenerate
relativistic gas. Mon. Not. R. Astron . Soc., 96, 780-4.
Penrose, R. ( 1965). Gravitational collapse and space-time singularities. Phys. Rev. Lett.,
14, 57-9.
Penrose, R. ( 1 969). Gravitational collapse : the role of general relativity . Riv. Nuovo
Cimento, 1, 252-76.
Prendergast, K . H . and Burbidge, G. R. ( 1968). On the nature of some galactic X-ray
sources. Astrophys. J., 151 , L83-8 .
Price, R . H . (1 972). Nonspherical perturbations of relativistic gravitational collapse.
Phys. Rev., D5, 24 1 9-54.
Rees, M. J. ( 1 984) . Black hole models for active galactic nuclei . Ann. Rev. Astron .
Astrophys., 22, 47 1-506.
Rindler, W. ( 1965). Elliptic Kruskal-Schwarzschild space. Phys. Rev. Lett. , 15, 100 1-2 .
Robertson, H . P . and Noonan, T . W . ( 1968) . Relativity and Cosmology, pp. 246-52.
W. B. Saunders : Philadelphia.
Robinson, D. C. ( 1 975). Uniqueness of the Kerr black hole. Phys. Rev. Lett., 34, 905--6 .
Robinson, I . , Schild, A . and Schucking , E. L. (eds.) (1965). Quasistellar Sources and
Gravitational Collapse, Editors' introduction, pp. i-xvii . University of Chicago Press :
Chicago.
Rosenfeld, L . ( 1973). Remarks in Seizieme Conseil de Physique Solvay, p. 174. Stoops :
Brussels.
Russell, H. N . ( 1 9 12). Relation between the spectra and other characteristics of the stars.
Proc. Amer. Phil. Soc., 51, 569.
274 W Israel
Russell, H. N. ( 1 914). Relations between the spectra and other characteristics of the
stars. Popular Astron., 22, 275-94, 33 1-5 1 .
Russell, H . N . ( 1 925). The problem of stellar evolution. Nature, 1 16, 209- 12.
Russell, H. N. ( 1939). Introductory remarks to the Paris meeting : Novae and White
Dwa1fs. Hermann : Paris ( 194 1). Reprinted in Schatzman ( 1958), p. 1 .
Russell, H . N . ( 1 944) . Note o n white dwarfs and small companions. Astron. J., 51 , 13.
Russell, H . N., Dugan, R. S . and Stewart , J. Q . ( 1 938). Astronomy, vol. 2 , pp. 9 18 ,
958-9. Ginn : Boston.
Ryle, M . ( 1950). Radio astronomy. Rep. Prog. Phys., 13 , 184-246.
Ryle, M . ( 1955). Meeting of Royal Astronomical Society, 13 May 1955. Observatory, 15,
104-8 .
Salpeter, E. E. ( 1964). Accretion of interstellar matter by massive objects . Astrophys. J.,
140, 796-9.
Sanchez, N. and Whiting, B. ( 1986). Quantum field theory and the antipodal
identification of black holes. Preprint.
Schatzman , E. ( 1958). White Dwarfs pp. 68-73, 166-70. North-Holland : Amsterdam .
,
Schmidt , M . ( 1 963). 3C273 : a star-like object with a large redshift . Nature, 191, 1040.
Schonberg, M . and Chandrasekhar, S . ( 1 942). On the evolution of main-sequence stars .
Astrophys. J., 96 , 16 1-72.
Schwarzschild, K. ( 1 9 16a). O ber das Gravitationsfeld eines Masses nach der
Einsteinschen Theorie. Sitzungsberichte Koniglich Preuss. Akad. Wiss. , Physik-Math .
Kl., 189-96. Translation in Detweiler ( 1982).
Schwarzschild , K. ( 1 9 16b). O ber das Gravitationsfeld einer Kugal aus inkompressibler
Fllissigkeit nach der Einsteinschen Theorie. Sitzungsberichte Koniglich Preuss. Akad.
Wiss. , Physik-Math. Kl., 424-34.
Sen, N. R. ( 1934). On the equilibrium of an incompressible sphere. Mon. Not. R . Astron.
Soc., 94, 550-64.
Shklovsky, J . S. ( 1953). On the nature of the Crab nebula's optical emission. Dokl. Akad.
Nauk. SSR, 90, 983 .
Shklovsky, J. S . ( 1955). On the nature o f the emission from the galaxy NGC4486 . In
Radio Astronomy, ed . H. C. van de Hulst, pp. 205-7. IAU Symposium No. 4, Jodrell
Bank, August 1955. Cambridge University Press : Cambridge.
Shklovsky, J. S. ( 1978). Stars: Their Birth , Life and Death . Freeman : San Francisco .
Simpson, M . and Penrose, R . ( 1973). Internal instability in a Reissner-Nordstrom black
hole. Int. J. Theor. Phys., 1, 183-97 .
Sky and Telescope ( 1 96 1 ). American astronomers report : First true radio star? Sky and
Telescope, 2 1 , 148 .
Sky and Telescope ( 1964). Dallas conference on super radio sources. Sky and Telescope,
27, 80-4.
Smith, F. Graham and Lovell, B . ( 1 983). On the discovery of extragalactic radio sou rces.
J. Hist. Astron. , 14, 1 55-65.
Souriau , J. M. ( 1965). Prolongements du champ de Schwarzschild. Bull. Soc. Math.
(France), 93, 193-207.
Sterne, T. E. ( 1933). The equilibrium theory of the abundance of the elements : a
statistical investigation of assemblies in equilibrium in which transmutations occur.
Mon. Not. R. Astron. Soc. , 93, 736-67 (see p . 750).
Stoner, E. C. ( 1929). The limiting density in white dwarf stars. Phil. Mag., 1, 63-70.
Stoner, E. C. ( 1930). The equilibrium of dense stars. Phil. Mag., 9, 944-63 .
Stromgren, B . ( 1937). Die Theorie der Sterninnem und die Entwicklung der Stemen .
Ergebnisse Exakt. Naturwiss., 16, 465-534.
Struve, 0 . ( 1960) . The problem of Cygnus A . Sky and Telescope, 20, 259-62 .
Struve, 0 . ( 1962). The Universe, chap. 5 . MIT Press : Cambridge, Mass.
Sullivan, W. ( 1979). Black Holes. Doubleday : New York.
Sullivan , W. T. III ( 1984) . The Early Years of Radio Astronomy. Cambridge University
Press : Cambridge.
Synge, J. L. ( 1 949) . Gravitational field of a particle. Nature, 164, 148-9.
Synge, J . L . ( 1950). The gravitational field of a particle. Proc. R . Irish Acad. , A53,
83- 1 14.
Synge, J. L . ( 1960). Relativity: The General Theory, p. 283. North-Holland : Amsterdam.
Szekeres, G. ( 1960). On the singularities of a Riemannian manifold. Pub/. Math.
Debrecen, 7, 285-30 1 .
Thompson, L. A . ( 1984) . High-resolution imaging from Mauna Kea : Cygnus A.
Astrophys . J., 279, L47-9.
Thorne, K . S. ( 1967) . The general relativistic theory of stellar structure and dynamics. In
High-Energy Astrophysics, vol . I I I , ed . C. DeWitt , E. Schatzman and P. Veron, p. 4 14.
Gordon and Breach : New York .
Thorne, K . S. ( 1970). Nonspherical gravitational collapse : does it produce black holes?
Comment. Astrophys. Space Phys., 2, 19 1-9.
Thorne, K . S. ( 1972) . Nonspherical gravitational collapse : a short review. In Magic
without Magic, ed . J. R . Klauder, pp. 23 1-58. W. H . Freeman : San Francisco .
Thorne, K . S . , Price, R . H . and MacDonald , D . A. (eds) ( 1986). Black Holes: The
Membrane Paradigm, chap . 1 . Yale University Press : New H aven.
Tipler, F . J . , Clarke, C . J . S . and Ellis, G. F . R . ( 1980) . Singularities and horizons - a
review article. In General Relativity and Gravitation, ed . A . Held , Vol . 2 , pp. 97-206 .
Plenum Press : New York.
Unru h , W. G. and Wald , R. M . ( 1982). Acceleration radiation and the generalized
second law of thermodynamics. Phys. Rev., D25, 942-58 .
Wali, K . C . ( 1982). Chandrasekhar vs. Eddington - an unanticipated confrontation.
Physics Today (October) , pp. 33-40.
Weisskopf, V . F. ( 1975) . Of atoms, mountains and stars : a study in qualitative physics.
Science, 1 87, 605- 12.
Weyl , H . ( 19 17). Zur Gravitationstheorie. Ann . Phys., 54 , 1 17-45.
Wheeler, J. A. ( 1955). Geans. Phys. Rev. , 97 , 5 1 1-36.
Wheeler, J. A. ( 1 966) . Superdense stars. Ann. Rev. Astron. Astrophys. , 4, 393-432 , see
p. 4 18 .
Wheeler, J. A. ( 1968). O u r universe : the known and the unknown. Amer. Sci. , 56, 1-20.
Whittaker, E. T. ( 1949) . From Euclid to Eddington, p . 124. Cambridge University Press :
Cambridge.
W olfendale, A. W . ( 1 986). Address on presentation of Gold Medal of the R.A.S. to
Ya. B. Zel'dovich. Q . J. R. Astron. Soc. , 27, 54 1-2 .
Zel'dovich, Ya. B . ( 196 1). The equation of state at ultrahigh densities. Sov . Phys. JETP,
14, 1 143-7 ( 1 962).
Zel'dovich, Ya . B. ( 1964). The fate of a star and the evolution of gravitational energy
upon accretion . Sov. Phys. Dok/., 9, 195-7 .
Zel'dovich, Ya. B . ( 1965) . Survey of modern cosmology. Adv. in Astron. Astrophys., 3,
24 1-379, see p . 335.
Zel'dovich, Ya. B . ( 1970). Generation of waves by a rqtating body. JETP Lett., 14,
180-- 1 .
Zel'dovich, Ya. B . and Novikov, I . D. ( 1964). Mass of quasi-stellar objects. Sov. Phys.
Dok/., 9, 834-7.
276 W Israel
Zel'dovich, Ya. B. and Novikov, I. D. ( 1965). Relativistic astrophysics. So v . Phys.

Uspekhi, 7, 763-8 8 ; 8, 522-77.
Zwicky, F . ( 1 938). On collapsed neutron stars. Astrophys. J., 88 , 522-5,.
Zwicky, F. ( 1939). On the theory and observation of highly collapsed stars. Phys. Rev . ,
55, 726-43.
Zwicky, F. ( 1 94 1). Neutron stars as gravitational lenses. Phys. Rev . , 59, 22 1 .
Zwicky, F . ( 1957). Morphological Astronomy, pp. 6, 255-6. Springer: Berlin.
Note added in proof
I am indebted to several colleagues, in particular J . Eisenstaedt, S . W .

Hawking, W . H . McCrea, W . G . Unru h and J . A . Wheeler, for their
constructive critical comments on the manuscript and suggestions, some of
which it has been possible to incorporate in the proof stages. A book by Kip
Thorne, which I understand is now actively in preparation, may be expected
to shape into a definitive history of relativistic astrophysics.
Gordon Baym kindly informs me that Rosenfeld 's dating and placing of
his conversation on neutron stars with Bohr and Landau · (see p. 224) was
probably a slip of memory, very understandable after a lapse of forty years.
At the time in question (February or March 193 2) none of the three
protagonists was in Copenhagen (Landau had moved to K harkov). The
_
conversation may have taken place during a meeting in the Soviet Union two
years later.
8
Astrophysical black holes
R . D . B LAND FORD
8.1 Introduction
The idea that stars might exist with surface escape velocities in excess of the
speed of light is at least two hundred years old (Michell , 1784 ; Laplace,
1795). However, it was not until the work of Chandrasekhar ( 193 1) that
there was a good reason that they ought to exist. As is well known ,
Chandrasekhar showed that white dwarf stars, supported by relativistic
electron degeneracy pressure had a maximum mass of 1 .4 M 0 (for a mean
molecular weight per electron of 2 ; c.f. also Landau , 1932). Somewhat
later, the discovery of the neutron led to the proposal that neutron stars also
exist (Landau , cited in Shapiro and Teukolsky, 1 983) , and the prescient
suggestion by Baade and Zwicky ( 1934) that 'supernovae represent the
transitions from ordinary stars into neutron stars ' . The maximum mass of a
neutron star was calculated (Oppenheimer and Volkoff, 1939) to be
0.75 M 0 . Although this is rather lower than the value calculated using a
modern equation of state, it verified the general conclusion that cold stars
were limited to a mass equivalent to (Gm; /hc) 3 12 nucleons.
,...,._,
These discoveries motivated Oppenheimer and Snyder ( 1939) to study

what would happen to aggregations of cold matter with mass in excess of
this limit. They understood the kinematic essentials of gravitational collapse
by analysing the radial infall of a pressure-free star. They showed that,
although the collapse would formally take an infinite time when viewed from
large distance, the time measured by an observer comoving with the star
would be finite as would also be the time until a distant observer would find
the star to be undetectably faint. (Interestingly, Einstein ( 1939) published a
paper in the same year analysing quasi-static clusters of masses moving on
circular orbits. He calculated that material particle orbits could not have
radii less than one and a half Schwarzschild radii, in Schwarzschild
278 R. D. Blandford
coordinates, and on this basis asserted that 'Schwarzschild singularities' do

not exist in physical reality. He believed that this would prove to be a general
conclusion.)
After a long interval, these ideas were elaborated by Wheeler and his
colleagues in the west (Harrison et al., 1965 ; Thorne, 1966) and Zel'dovich
and his colleagues in the Soviet Union (Zel'dovich and Novikov, 1965,
1966) during the 1960s. Wheeler ( 1966) in particular emphasised that the
very possibility of a physical singularity (as opposed to the coordinate
singularity at the Schwarzschild radius) posed a strong challenge to
theoretical physics and declared that it was 'difficult to name any situation
more enveloped in paradox than the phenomenon of gravitational collapse'.
The singularity theorems by Penrose ( 1965) and Hawking ( 1965) implied
that singularities could not be avoided by minor departures from spherical
symmetry. Apparently, none of the 'accidents' desired by Eddington ( 193 5),
in his famous attack on the idea of gravitational collapse, would intervene to
save a collapsing star. (It is interesting to speculate upon whether Eddington
would have retained his prejudices against the formation of gravitational
singularities had he understood the pri nciple of 'cosmic censorship'
(Penrose, 19�9).)
These developments merged with parallel advances in studies of the
solutions of the Einstein equation with the publication (Kerr, 1963) of the
stationary axisymmetric metric that now bears his name. However, as
described in Carter ( 1 979), ten years were to elapse before the work of lsrael,
Carter, Robinson and Hawking showed that the Kerr metric almost
certainly described the only uncharged , asymptotically flat vacuum
spacetime that would remain after gravitational collapse, thus vindicating
Wheeler's famous 'no hair' conjecture. It is perhaps not surprising that in
the middle of these discoveries, 'collapsed stars', with their implied strong
connection to their stellar parents should become the 'black holes' of
Wheeler ( 1968); a name which embodies not just the absorbing properties of
an event horizon, but the independence of the structure from the history of
the collapse.
The theoretical framework for carrying out investigations of black hole
astrophysics was therefore largely established in time for the discovery of
galactic X-ray sources by Giacconi et al. ( 1962) and quasars by Schmidt
(1963). However, less well developed were ideas about how the presence of
black holes might be inferred by astronomers. The reason is not hard to
uncover. In order to address this issue, the theoretical astrophysicist must
consider ill-posed problems in radiative transfer, magnetohydrodynamics,

plasma physics, stellar dynamics and stellar evolution, not to mention
the vagaries of interpreting astronomical phenomenology. This is a
qualitatively different type of research from formal investigations of the
physical properties of an isolated black hole. In the absence of any strong
observational stimulus, such research is inevitably highly uncertain and
speculative. Nevertheless, there were some prophets and it is with the
development of their ideas leading up to the present observational
position on the existence of black holes that we are concerned in this
article.
Black holes, by definition, are structures involving strong field general
relativity and can only be properly described in these terms. Nevertheless, it
is a feature of (and perhaps a sad commentary on) current progress in black
hole astrophysics that virtually all the physical processes that are likely to be
observable can be described with adequate precision using pre- general
relativity (although not entirely Newtonian) models. Indeed in most of the
relevant astronomical literatu re metrics, horizons, etc. are avoided and
calculations are performed in a flat space terminated when necessary at the
Schwarzschild radius. Much of the discussion that follows will be couched in
this language, not as a reactionary attack on general relativity, but instead
as a means of emphasising that astronomical observations of black holes are
not very promising ways to verify that general relativity is the correct theory
of gravitation. Solar system tests and the first binary pulsar PSR 1 9 1 3 + 16
are far more important in this respect.
Much of the subject matter encompassed by this article has been more
completely reviewed elsewhere. In particular, in an earlier volume
celebrating the centenary of Einstein's birth, Blandford and Thorne ( 1 979)
reviewed this topic as it stood in 1978. In this volume, I will therefore just
describe the key observational and theoretical discoveries, in keeping with
the historical theme of the present volume and summarise comparatively
recent developments.
In the following section, I describe some physical effects believed to be
important for astronomically occurring black holes. In Section 8.3, I
summarise the observational evidence for the existence of black holes in the
two environments where we have good reason to believe that they can be
found , i .e. active galactic nuclei and accreting binary X-ray sources, as well
as list some of the more speculative possibilities. Prospects for the near
future are briefly summarised in Section 8.4.
280 R. D. Blandford
8.2 Black holes in astrophysics
8.2 . 1 General remarks

Although some physicists have regarded the quite genuine difficulties that
the creation of a singularity poses for physics as somehow providing a
teleological argument against the existence of black holes, and rather more
observational astronomers have retained a perfectly proper scepticism
about the evidence that has been advanced for their presence, most
astrophysicists now see black hole formation as a natural and inevitable
consequence of stellar and galactic evolution. No longer is it considered
obligatory to insert adjectives like bizarre or exotic in popular accounts. The
description of the usual foolhardy astronaut , although strange by everyday
standards, now seems qualitatively similar to the gedanken experiments of
special relativity. Besides, the principal properties of black holes are much
easier to communicate than, for example, the behaviour of quarks and
gluons. In short, astronomers and physicists have got used to the idea.
However, none of this familiarity removes the obligation to make good
our hunches and demonstrate, beyond all reasonable doubt, that black
holes really do exist. Here progress has been distressingly slow and no better
individual candidate than Cygnus X- 1 , first discussed in 1972, has been
forthcoming and to date only two more good cases can be claimed. This
contrasts with the observational position on neutron stars, where, following
the discovery of the first radio pulsar by Hewish et al. in 1967 , we now have
nearly 500 examples of neutron stars (Lyne, Manchester and Taylor, 1985).
(Although, even here, the possibility that we might be observing quark stars
cannot be completely dismissed (Alcock , 1986).)
The incentives for proving that black holes are really present in binary X
ray sources, active galactic nuclei and perhaps elsewhere are twofold .
Firstly, measurement of their frequencies, masses and spins would certainly
clarify our understanding of the endpoints of stellar evolution, and rather
more remotely, the early phases of galactic evolution. Secondly, and of more
interest to physicists, they offer the opportunity of observing some of the
distinctive features of strong field gravity. Here, we should not be too
optimistic. The 'no-hair' theorem condemns us to observing the gas orbiting
a black hole and , unlike the case with a magnetised neutron star, it is
extremely difficult to imagine making a clean, quantitative measurement of
a relativistic effect that is not hopelessly confused by the erratic
magnetohydrodynamic behaviour of the gas. Indeed , it is one of the peculiar
features of this subject, that we find it fairly easy to include the twentieth
Astrophysical black holes 28 1
century physics o f general relativity and quantum mechanics i n our

calculations; it is the nineteenth century subjects of fluid mechanics and
electromagnetic theory that cause the problems. Black holes may provide a
useful quantitative test of general relativity. A corollary of this is that most
serious modelling can , and is performed , using just 'Newtonian' physics. In
the spirit of this volume, we now describe some relevant properties of
relativistic black holes in Newtonian language.
8 .2 .2 The gravitational field around a black hole

As is well known , general relativity causes the gravitational force around a
black hole to become infinite at a finite radius, the black hole horizon . In the
case of a non-rotating hole, this is usually described by the Schwarzschild
metric. However, this behaviour can be imitated by using a Newtonian
potential </> = - m/(r - 2m), setting G = c = 1 (e.g . Paczynski and Wiita,
1980 ) . The sum of the gravitational potential and the centrifugal potential
define an effective potential for radial motion Cf> = </> + 1 2/2r 2 , where l is the
specific angular momentum. This effective potential has a minimum if
l ?:- (2 7/2)1 ' 2m. The most bound stable circular orbit has an angular
momentum l = (27 /2)1 ' 2 m, a radius r = 6m and a binding energy 1/24, rather
similar to the Schwarzschild values. Cold gas in orbit about a black hole will
form a thin accretion disk extending down to 6m, presumably evolving
under internal viscous stresses. The thickness of the disk will be given by
balancing the vertical gravitational acceleration 9 z = - mz/r(r - 2m) 2 with
the internal pressure gradient divided by the gas density. If the gas is
isothermal , with sound speed s, the disk will have a Gaussian density profile
with half width s(r - 2m)(r/m)1 ' 2 . Again , this expression is similar to its
relativistic counterpart (e.g . Novikov and Thome, 1 973).
Now, if the infalling gas is unable to cool fast enough, either by accreting
so rapidly that the gas traps the escaping radiation, or so slowly that two
body processes cannot produce photons fast enough, then a thick accretion
disk or torus may be produced . In this case, gas moving in the equatorial
plane of the torus can orbit stably within r = 6m, held in place by a pressure
gradient. If, for example, the gas has a constant specific angular momentum
l, the torus will steadily inflate until its innermost equatorial orbit coincides
with the saddle point in the effective potential where a cusp will be found. If
the torus tries to inflate further then the gas will simply pour through the
cusp onto the black hole. The gas at the cusp will be bound (i .e. have Cf> < 0)
as long as l > 4m and the radius of the associated marginally bound circular
282 R. D. Blandford
orbit is 4m. The surface of marginal binding is given by = 0 or the

paraboloid
(p - 4m)(p + 4m)
z = ------- (8 . 1)
8m
where p is the cylindrical radius. This surface defines a funnel which may
possibly collimate outflowing jets (Lynden-Bell, 1978 ; Fig . 8 . 1 ) . If the gas is
more tightly bound than this then it will form a torus in orbit about the
black hole. Again very similar results are produced using a relativistic
analysis (Abramowicz, Jaroszynski and Sikora, 1978).
This simple Newtonian model can only describe a non-rotating black
Fig . 8 . 1 . Contours of constant effective potential around a black hole

described using a model Newtonian potential , </> - m/(r 2m) (Distances
=
-
.
are measured in units of m and only negative contours are shown.) The
specific angular momentum is chosen to have a constant value l = 4m so that
the isobars coincide with the equipotential surfaces and the contour passing
through the cusp has zero binding energy. Gas can fill up the region occupied
by the contours outside r = 4m, and may pour through the cusp onto the
black hole. The ze�o energy surface defines a pair of fu nnels which may be
responsible for channelling some of the outflow and beaming some of the
radiation in bright quasars and Seyfert galaxies.
hole. However, we can describe the gravitational far field around a spinning
black hole using an electromagnetic analogy and decomposing it into a
gravitoelectric field g (the usual Newtonian gravity) and a gravitomagnetic
contribution H. The Einstein equations for g, H for material of density µ and
velocity v � c approximates to a form strikingly similar to Maxwell's
equation ; the geodesic equation for the motion of a freely falling particle
becomes equivalent to the Lorentz force law;
V · g = - 4nµ,
V ·H = O '
V x g = O,
V x H=4 [ - 4 xµv + ��].

dv
v x H,
dt = g +
(8 .2)
(e.g . Braginsky et al., 1977, where the equations are actually given to second
order in v ) . g and H can be derived from scalar and vector potentials,
respectively ; g = - V¢, H = V x y where ¢, y are related to the metric tensor,
9a.p ' by ¢ = ( 1 + g 00 )/2 , 'Yi = g 0i . Note the minus sign in the 'Poisson' and
'Ampere' equations expressing the attractive character of gravity and the
extra factor of 4, which relates to the spin-2 nature of the gravitational field.
Now, from our electromagnetic experience, we can infer immediately that
at a distance r � m , where the space is approximately flat, a hole with spin
angular momentum S will be surrounded by a gravitoelectric field
g = - mr/r 2 where the gra vitoelectric charge is the hole's mass m . If we use
the classical gyromagnetic ratio, we identify S/2 as the gravitomagnetic
dipole moment. Now if we use the fact that the gravitomagnetic field is - 4
times the usual dipolar field, w e obtain an expression for the dipolar
gravitomagnetic field surrounding a spinning black hole,
H = 2 [S - 3(S r) r]/r 3
· , (8 .3)
g and H communicate information to a distant observer about the two
parameters m , S which characterise an uncharged black hole.
We can now use these fields to calculate the rate of precession of a ring of
gas in inclined orbit about a spinning black hole with specific angular
momentum I . Pursuing our electromagnetic analogy, a torque per unit mass
of r x (v x H) will act upon elements of the ring . Using the formula for
dipolar gravitomagnetic field and averaging over azimuth for a small angle
of inclination gives a mean torque per unit mass of 2(S x l )/r 3. The
284 R. D. Blandford
gravitomagnetic or Lense-Thirring ( 19 1 8) precession frequency is therefore

given by 2 S/r3 . .
The importance of this precession for astrophysical black holes is that it

may provide a mechanism for driving an orbital accretion disk into the
equatorial plane of a spinning black hole and so communicating the
gyroscopically stable spin direction of the hole to the direction of the
outflowing jets (Bardeen and Petterson , 1975; Rees, 1978). Adjacent rings of
a gas within an inclined accretion disk should precess at different rates and
friction between them should drive the innermost rings into the equatorial
plane. If we adopt the criterion that this mechanism is efficient when the
precession period is shorter than the inflow time (but see Kumar and
Pringle, 1985), we conclude that the disk should lie in the equatorial plane
out to a radius r8 p (S/SmaJ213(vr/v) - 213m, where Smax = m2 is the maximum
""
allowed spin angular momentum of the black hole and vr is the inflow speed .
(We discuss this further below.) A corollary is that the hole may change its
spin substantially after it has increased its mass by a fractional amount
(S/SmaxHm/r8p) 1 12. Some extragalactic radio sources show jets that bend on
both sides of the nucleus with a pronounced 'inversion' or 'S' symmetry (see
Fig. 8 .2). If jets are launched along the spin axis of the hole, then this type of
distortion may be related to changes in the spin direction of the hole.
Fig. 8 .2. 'Inversion symmetric' extragalactic radio source associated with the
galaxy NGC326 (marked with a + ) . The apparent S-type symmetry of the
radio contours might be caused by a systematic change in the orientation of
the spin axis of a central black hole. (R. Ekers, personal communication.)
NGC326
I' l'
Dec
0 0
-I' 0 lI
RA
Astrophysical black holes 28�
A related precessional effect is called geodetic precession. Consider a

gyroscope, e.g. a smaller spinning black hole in orbit about a massive body
(e.g . a large black hole) . The motion of the gyroscope through the
gravitoelectric field induces a gravitomagnetic field H = - v x g in its frame.
Again using our electromagnetic knowledge, we find that the gyroscope will
precess with an angular frequency ns o = - H/2 = ml/2r 3 . This precession is
the analog of spin-orbit coupling in a hydrogen atom. There is no Thomas
term to be included because the gyroscope is freely falling rather than being
accelerated relative to a local inertial frame by a non-gravitational force.
(When the massive body is also spinning , there will be an additional
contribution to the precession rate caused by its gravitomagnetic field
nGM = - H/2 = [3(S · r) r - S/r 3 ] . (8 .4)
Usually, this will be small compared with the geodetic prece s sion.)
However, there is a contribution to the precession rate attributable to the
space curvature. Using an embedding diagram for the Schwarzschild metric
(e.g . Misner, Thorne and Wheeler, 1973) it can be seen that the gyroscope
will precess through an additional angle ""' 2nm/r per orbital period . Adding
this to the spin orbit precession gives a total geodetic precessional frequency
of
ngeo = 3!1s o = 3ml/2r 3 . (8 .5)
This formula has been derived assuming that the mass ratio is small . When
the reduced mass of the two body system, µ is comparable with m then the
total precession rate becomes
n geo = (3m + µ)l/2r 3 ' (8 .6)
where I and r now refer to the relative orbit. (Barker and O'Connell , 1975).
In the limit that the gyroscope becomes much heavier than the first body of
mass m, the mean torque acting on it becomes equal to 2(1 x S )/r 3 , equal and
opposite to the Lens-Thirring torque computed above, as it should.
These arguments and the relationship to the proper relativistic
calculations are discussed at length in Thome, Price and MacDonald
( 1 976).
Again there may be observational consequences of geodetic precession if,
as proposed by Begelman , Blandford and Rees ( 1 980) , black hole binaries
can form in the nuclei of galaxies. Typical precessional periods
""' I0 4 (a/0.0 1 pc) 512 (m/ 108 M 0 ) - 312 yr for typical orbital radii a "' 0.0 1 pc
and masses m ""' 108 M 0 are deduced . Geodetic precession may also be
responsible for those radio jets that exhibit pronounced inversion
symmetry.
286 R. D. Blandford
8 .2 .3 Electromagnetic effects
Another aspect of black hole astrophysics that is amenable to a classical or
more specifically 'Maxwellian' interpretation is the interaction of a black
hole with a magnetic field. Although charged , Kerr-Newman black holes
are irrelevant to astronomy, uncharged Kerr holes interacting with
magnetic fields supported by currents external to the horizon may be highly
relevant. Some intuition about this interaction can be obtained using a
sequence of four thought experiments.
Firstly, if a Schwarzschild black hole is placed in a uniform electric field,
then , after the· transients have decayed, a modified static field will remain in
which the physical components of the electric field (as seen by an observer
hovering over the horizon) becomes radial at the event horizon (e.g. Press,
1972). The observer might therefore think that the black hole is behaving
like an electrical conductor and he can introduce a fictitious surface charge
density on the horizon consistent with Gauss' theorem. This representation
of an event horizon as a 'quasi-Newtonian' surface embodied with
conductivity, as well as viscosity, surface pressure, momentum etc.
(Damour, 1978 ; Znajek, 1978) is termed the membrane paradigm and has
been extensively explpred by Thorne and colleagues {Thorne, Price and
MacDonald, 1976).
We ought to clarify what is meant by electric and magnetic fields at this
stage. One way to do this is to define a set of fiducial observers. These
observers will be at rest in Schwarzschild coordinates if the hole is non
rotating and coincide with the 'zero angular momentum' observers in the
spacetime of a rotating hole (Bardeen , 1973). The electric and magnetic
/
fields that we are talking about are the ones that they would measure . In the
absence of free charges, these fields are solenoidal on the 3D hypersurfaces of
constant Schwarzschild or Boyer-Lindquist time. We can therefore define
electric and magnetic field lines (see Fig. 8 .3).
In a second thought experiment, let a cloud of magnetised plasma fall into
a black hole. As it approaches the horizon, the cloud will be tidally distorted
and the electromagnetic field will vary rapidly. After the plasma has crossed
the horizon, electromagnetic fields will linger in the neighbourhood of the
horizon for roughly a light crossing time before decaying to leave behind a
'bald' black hole (consistent with the no-hair theorems) . This tells us that if
we regard the event horizon as possessing a conductivity, its value cannot be
infinite. In flat space, a sphere of radius A , and electrical resistance R, will
lose magnetic field in a time A /Rc 2 • Equating this to the light-crossing
,...,
time ,..., A./c, tells us that the resistance of the hole is RH ,..., 1/c = 30 n .
More precisely, if we try to solve Maxwell's equations in the curved

spacetime surrounding a Schwarzschild black hole, we find that the
boundary conditions at the horizon dictate that the horizon must appear to
have a surface resistivity of 377 Q . Consider an observer freely falling from
infinity with speed v = (2m/r) 1 1 2 relative to a stationary observer just
hovering above the horizon. Let the Lorentz factor be y = ( 1 - 2m/r) - 1 ' 2 as
usual . The electric and magnetic fields measured by the infalling observer,
E', B', must remain finite near the horizon. However the stationary observer
will see fields that look just like an electromagnetic wave propagating
radially inwards . That is to say both the electric and the magnetic field will
be equal , mutually perpendicular, and parallel to the horizon to a fractional
accuracy y - 1 . This is akin to the Weizsacker-Williams method of virtual
-
quanta. Let the components of electromagnetic field measured by a fiducial

observer hovering just above the horizon be E rr • B rr · The magnitude of these
field components, (E r • B r ) y max(E', B') will diverge as the hovering
-
observer approaches the horizon. The Poynting energy flux measured by

this observer will diverge ex y 2 (one factor of y for the blue shift of the
photons, another to accommodate the change in photon rate). However, we
can define renormalised horizon fields
E H , B H = lim E r h , B r h , (8 .7)
r - 2m
Fig . 8 . 3 . An isolated Schwarzschild black hole is immersed in a uniform
electric field . (a) lfwe plot the electric field lines in Schwarzschild coordinates
(in which a finite interval of proper radius close to the horizon corresponds to
a very small interval of coordinate radius), the field lines appear to remain
straight . (b) If we transform the radial coordinate to proper radius, then the
field lines curve and cross the horizon normally. Observers hovering just
above the horizon and falling through it radially will both measure a field
perpendicular to the horizon . (Adapted from Thorne et al. , 1976.)
(a) (b)
288 R. D. Blandford
lying in the horizon, which remain finite. (The radial field components
remain finite without any renormalisation.) Note that
EH = E' + r x B',
BH = B' - i x E' , (8 .8)
verifying that EH, BH are equal in magnitude and perpendicular (see Fig .
8 .4).
Furthermore, the Poynting energy flux crossing the horizon, as measured
by an observer at infinity, will be smaller than that measured by the
Fig. 8 .4. (a) When an isolated black hole is penetrated by electric field lines,
we can imagine that these terminate on surface charges located just above the
horizon (i.e. on an imaginary surface called the stretched horizon). The
associated surface charge density is given by E .i /4n. (b) A renormalised
component of tangential magnetic field B H just outside the horizon can
likewise be thought of as terminating at a surface current density J H
r x BH/4n. (c) The surface boundary condition discussed in the text implies
=
that there is a renormalised electric field parallel to the horizon EH = 4nJH ,

equivalent t o saying that the surface resistivity i s 377 Q . (d) When currents
flow across the horizon, charge conservation can be thought of as being
satisfied by surface current flowing in the horizon and surface charge density
accumulating there. (Adapted from Thome et al. , 1 976.)
(a) Gauss's law (b) Ampere's law
(c) Ohm's law (d) Charge conservation

stationary observer by two redshift factors and given by EHBH/4n = E�/4n

which will also be finite. Its integral over the area of the horizon will give the
rate of increase of hole mass. Now if we imagine that there is no tangential
magnetic field inside the horizon, we must terminate an exterior tangential
magnetic field with a surface electric current in the horizon . J 8 = B 8/4n
perpendicular to the tangential magnetic field just outside the horizon.
Using our boundary condition EH = BH, we derive the constitutive relation
18 = S8EH where SH , the surface resistivity, equals 4n - 377 Q. Not
surprisingly, the resistance of the horizon approximately equals the
impedance of free space. These results carry over to the Kerr metric. For
more details of those arguments, see Carter ( 1979) and Thome, Price and
MacDonald ( 1 976) and references therein.
In a third experiment (Damour, 1978), a potential difference V is applied
between the equatorial and polar regions of a Schwarzschild black hole
using a battery of internal resistance much smaller than R H . Provided that
both signs of charge are available to cross the horizon, current I will flow
around the circuit and through the hole. Endowing the hole with resistance
RH implies that I "' V/RH and that the hole will grow at a rate m "' V 2/RH.
Now immerse the hole and circuit in a uniform field B0 parallel to the
polar direction. There will be a Lorentz force acting on the hole and this will
produce a torque "' mJ8B0 "' VB0/R8 which will start the hole spinning just
like a simple electric motor. In this case there is a flux of electromagnetic
angular momentum as well as energy crossing the horizon (see Fig. 8.5).
The fourth thought experiment is the one that concerns us most directly
and is the converse of the last one (Blandford and Znajek , 1977). Suppose we
have a spinning black hole immersed in a uniform field B0 , parallel to the
spin axis. If the hole is surrounded by vacuum, then because the hole is a
conductor, it will act as a unipolar inductor and a potential difference V"'
nH m 2 Bo "' n H ' where is the flux threading the hole, will be induced
between the poles and the equator and nH is the angular velocity of the hole.
(QH must not exceed Qmax = l/2m, the limiting angular Velocity for a hole of
total mass m.) As the hole is a conductor, the electric field measured by
fiducial observers hovering just above the horizon and orbiting with an
angular velocity very close of 08 will be radial and E 8 = 0. Our boundary
condition then implies that B8 = 0 as the full electromagnetic solution
confirms (Wald, 1974). Just as we can introduce a fictitious surface current
density to terminate a tangential magnetic field, so, in this example, we can
introduce a quadrupolar surface charge density to terminate the radial
electric field . It is interesting that the EMF can be thought of as arising from
290 R. D. Blandford
an interaction between the hole's gravitomagnetic potential and the external

magnetic field (e.g. Thorne, Price and MacDonald, 1976; Fig. 8.6).
There will be no change in the mass or the angular momentum of the hole
as long as it remains an open circuit and no current flows. However, if we
now connect the pole to the equator via a load with large resistance RL � RH ,
then a small current will flow of magnitude 1 - V/RL. The power that is
dissipated in the resistive load, V 2IRL ,..., n�;RL is drawn from the spin of the
hole. Now, as originally shown by Christodoulou ( 1 970), the mass m of a
spinning hole may be decomposed into an irreducible part, mi, and a spin
energy. This spin energy can , in principle, be extracted by Penrose-type
processes such as the one we are analysing here. m is related to mi by
m = mi(l - 4mf n � ) - 11 2 , (8 .9)
a formula similar to that appropriate for rectilinear motion with the
Fig. 8 .5. A potential difference V is applied across a Schwarzschild black hole

so that a current I "' V/RH will flow, inward at the pole and outward at the
equator. The circuit can be regarded as being completed by a surface current
J H "' I/2nm. Associated with this surface current is a horizon electric field
EH "' IR H /2nm and a perpendicular magnetic field BH = EH. There is an
2 2
inwardly directed Poynting flux N"' 1 RH/2nm and the hole gains mass at a
2 2
rate m "' 2nm N "' I R H . Now if an external magnetic field B0 is applied as
shown , there will be a torque per unit area "' mJHBo which will start the hole
spinning. (Adapted from Thorne et al., 1976.)
important difference that the 'velocity' 2mi Q H is limited to 2 - 1 1 2 rather than

unity.
The way in which power is extracted is quite subtle. The boundary
condition dictates that a fiducial observer just above the horizon will detect
Fig. 8.6. (a) W hen an isolated hole spins (with angular velocity QH), parallel
to an external magnetic field B0, an electric field and an apparent
quadrupolar surface charge density will be induced . The potential difference
between the pole and the equator is V- QH B0m. (b) W hen the pole and the
equator are connected via a load of resistance RL, a current will flow and
power will be dissipated in the load. This power derives from the spin of the
hole. (Adapted from Thorne et al. , 1976.)
B
I
)
I
/
I
(a)
292 R. D. Blandford
a small, inwardly-directed Poyn ting flux, diverging as he approaches the

horizon as our boundary condition requires. However, he will also measure
an outwardly directed electromagnetic angular momentum flux. Now, if we
look at this from the point of view of a distant, non-rotating, observer
(inserting the appropriate redshift factors), the energy and angular
momentum fluxes combine so that the conserved energy flux at infinity will
be outward. All but a fraction ""' RH /RL of the reducible mass can be
extracted in this manner.
As the value of the resistive load is reduced, the rate of power extraction
will increase oc RC 1 . However, when RL ::::;; RH , we must take into account the
internal resistance of the battery, and the power extracted will be
""' V2RdRL + RH) - 2. For a given emf, this is maximised when the load
resistance matches the internal resistance of the black hole. (This is a
circumstance which may occur in practice if the load is actually a relativistic
electromagnetic wind in which the electric and magnetic fields are nearly
equal.) In the limit when the hole is effectively short-circuited , the current
will be I ""' V/RH and the hole will slow down at more or less constant total
mass.
A black hole surrounded by a magnetised accretion disk may therefore act
as a unipolar inductor. Up to 29 per cent (for nH = nrnaJ of its mass is
reducible and this may be dissipated in a sufficiently resistive load . The
magnetic field , supported by currents in the surrounding disk, should be
reasonably stationary and axisymmetric as the Bardeen-Peterson effect
drives the disk into the equatorial plane of the hole. Instead of the wires
invoked above, currents should flow into and out of the hole through the
magnetosphere and connect at large distance from the hole where they
presumably accelerate charged particles (the load). As particles have to flow
in at the horizon and outward at large distance from the hole, they must
have a source in the magnetosphere. If cross field diffusion is too slow to
accomplish this, then, under conditions envisaged within galactic nuclei ,
electrons and positrons can be copiously created by y-ray collisions.
The electromagnetic structure of a stationary, axisymmetric
magnetosphere is conceptually quite simple. If the particles in the
magnetosphere have negligible inertia, then the electromagnetic fields
must be force-free, i.e. pE + j x B = 0. This implies that the axisymmetric
magnetic surfaces must be equipotential and, by Ferraro's law have a well
defined angular velocity w specifying the frame in which the electric field
vanishes. Stationarity of the magnetic field requires that the electric field be
poloidal which in turn, through the force-free condition , implies that the
current flows along the magnetic surfaces. These surfaces therefore replace
the wires used in the circuit analysis.
Now decompose the magnetosphere into many nested elementary circuits
in which the current flows along adjacent magnetic surfaces and connects
across an annular ring of the horizon (resistance L\R8) and a part of the load
(resistance L\RL) . When the load supplies most of the resistance in the
elementary circuit (L\R8 � L\RL), the field lines will move with the hole
(w "" !18). When the load resistance is negligible (e.g . if it is a highly
conducting accretion disk), the magnetic field lines will be frozen into it and
will rotate with its angular velocity nL (e.g . MacDonald and Thome, 1982).
If we transform into a frame corotating with the magnetic field lying
between two magnetic surfaces, with angular velocity w, the potential
difference across the annular ring of horizon is L\ V8 = (!18 - w)L\<l>/2n, where
L\<l> is the magnetic flux between the two surfaces. Similarly, the potential
difference across the load is L\ VL = (w - !1dl\<l>/2n. Now current
conservation requires that I = L\ V8/L\R8 = /.\ VL/L\RL or
w - QL L\RL
(8. 10)
!18 - w L\R8
The magnetosphere acts as a sort of clutch that couples the hole to the load
and rotates with a c9mpromise angular velocity w = (!18f\RL + nLL\R8)/
(L\RL + L\R8). As it has no inertia itself, it transmits a torque L\G = I L\<l>/2n =
I 2 (L\R8 + L\RL)/(!18 - n L ). This torque does work at a rate - f\GQ8 on the
hole, increasing its spin energy at this rate . It also does work at a rate f\GQL
on the load , increasing its spin energy. The sum of these two powers, which
of course must be negative, represents dissipation at a rate 1 2 L\R8 in the hole
plus 1 2 L\RL in the load.
Next, let us idealise the problem further by imagining that the hole is
rotating with a moderate angular velocity (!18 "" 1/8m) and that low latitude
magnetic field lines connect to the inner parts of the disk which is an
excellent conductor and rotates more rapidly than the hole with angular
velocity no > nH. Intermediate latitude field lines connect to the disk at
large radii where the disk is rotating less rapidly than the hole. Finally, high
latitude field lines are open and accelerate a relativistic wind.
Under these conditions, the black hole may evolve in several different
ways. The disk-connected flux will exert a torque on the hole given by an
integral over the horizon
GD =
J (n o - nH)r�Bn d
SH
<l> , (8 . 1 1)
294 R. D. Blandford
where 2nrH is the circumference of the horizon at that latitude and Bn is the
normal component of magnetic field at the horizon . Clearly G0 can have
either sign and in particular, it is possible for the disk to spin up the hole in
an analogous manner to the spin up in pulsating X-ray binaries. The mass of
the hole will increase at a rate that is the sum of the ohmic dissipation and
the net work done on it by the disk torque. Conversely, a rapidly spinning
hole may actually inhibit disk accretion if it is magnetically coupled to the
inner edge of the accretion disk .
The open field lines can also change the mass and the spin of the hole.
However, the boundary conditions at large distance are problematical . If we
try to match the outflow onto a wind with slow speed "" v , then the effective
impedance will be RL "" V � RH. Energy will only be extracted from the hole
with low efficiency in this case. However, if the outflow is relativistic then we
can have a good impedance matching and roughly half the spin energy can
be liberated. Unfortunately, it is hard to be more quantitative than this, as
we have to match the magnetospheric solutions onto an outflowing solution
and thus determine the value of w on the field lines.
In an important extension of these ideas, Phinney ( 1983) (cf. also
Damour, 1975; Carter, 1979) has replaced the force-free equations with
relativistic M HD theory. This allows particles, as well as electromagnetic
fields to transport energy and angular momentum. In the limit that the fluid
motion dominates the magnetic stresses, the Penrose process is recovered .
Phinney argues that, although the particle stresses may be small , the
requirement that the outflowing plasma pass through the necessary Alfvenic
and magnetosonic critical points to give a supersonic wind may actually
determine w and therefore the efficiency of energy extraction , independent of
the details of the dissipation in the load . Another aspect of the theory which
must be treated self-consistently is the shape of the magnetic surfaces . These
will be determined by imposing poloidal force balance and , if we are given a
prescription for fixing w, by the distribution of toroidal current in the
accretion disk.
Electromagnetic extraction of energy from a spinning hole is a strong
candidate for powering the jets of relativistic plasma that are observed in
extragalactic radio sources. It may also be relevant to stellar mass black
holes in mass transfer binary systems.
8 .2 .4
Accretion onto black holes
.
As this topic has been extensively reviewed elsewhere, I shall just consider a
few recent developments.
Perhaps, the simplest type of accretion to treat is spherically symmetric

accretion onto a Schwarzschild black hole, and not surprisingly, this still
attracts the most theoretical interest. Like, the problem of the axisymmetric
radio pulsar, it is deceptively simply posed, yet full of subtleties and
ultimately of little relevance to interpreting the observations. Nevertheless,
it has proved to be an important testbed for theoretical ideas with the bonus
that numerical computations are relatively easy to carry out. Recent
discussions include Krolik and London ( 1983), Schultz and Price ( 1985) and
Colpi, Maraschi and Treves ( 1984).
Gas accreting onto a black hole in a binary system or an AGN will almost
certainly have enough angular momentum to encircle the hole and
presumably to form some sort of disk or torus around it . Close to a spinning
hole, we expect the Lense-Thirring precession to drag the gas into the
equatorial plane. Unfortunately, almost everything we want to know to
interpret the observations depends upon the effective viscosity and we are
still no nearer to having a good physical understanding of this (e.g . Pringle,
1 98 1). Several different modes of disk accretion have been considered. When
the rate of mass accretion is very low compared with the Eddington rate, the
accreting gas may be unable to cool on the inflow timescale. The ions will
therefore acquire a temperature comparable with the virial temperature and
form a thick ring or torus. This can only happen if the electron temperature
is much lower than the ion temperature. These conditions are probably
optimal for the production of y-rays and electromagnetic extraction of
energy from the hole. In the case of AGN, the radio galaxies (which are able
to produce more power in the form of relativistic plasma than apparently in
the form of optical and UV photons) have been associated with this case. If
this is correct then it might be interesting to search for some galactic
counterparts.
At moderate rates of accretion, up to the Eddington rate, a fairly thin disk
is expected to form . The structure of the outer disk at least is probably fairly
independent of the nature of the central object and we can probably study
them best in cataclysmic variable stars comprising a low mass 'red' star and
a white dwarf. Indications are that the viscosity can be fairly large,
compatible with the empirical rJ, approximation in which the shear stress is
proportional to the pressure with a constant of proportionality ......, Q.0 1 - 1
(Shakura and Sunyaev, 1 973). As the inner parts of accretion disks can be
radiation-pressure dominated, an important issue is whether to use the gas
pressure or the radiation pressure in applying this prescription (Sakimoto
and Coroniti, 1 98 1). This can have a profound effect on the properties of
296 R. D. Blandford
disks around massive black holes in AGN. Limit cycle behaviour can be
observed in accretion disks and may be explained by a variety of different
mechanisms leading to episodic accretion (e.g. Lin et al. , 1985 ; ·s hields et al.,
1986). There is no guarantee that the accretion will be steady. In addition, as
the disk is differentially rotating, it may be able to generate magnetic fields
by some sort of dynamo process (e.g. Pudritz, 1983) as well as convect them
inwards. These fields can energise a corona where non-thermal radiation
I
mechanisms will predominate. This is particularly important in the context

of quasars which seem to produce almost as much power through radio,
infra-red , and X-ray 'non-thermal' channels as through 'thermal' emission
in the UV with an effective temperature appropriate to an emitting area
roughly 10 Schwarzschild radii across. Recent studies (e.g. Svennson, 1986)
have explored the possibility that an electron positron plasma may be
created by y-ray collisions, and that this might be responsible for
reprocessing the outflowing 'thermal' emission as well as producing the
'non-thermal' emission .
An important difference between accretion disks i n binary X-ray sources
and AGN is that in the former case the angular momentum of the inflowing
gas which is generally assumed to be transported outwards through the disk
by viscous stresses can be extracted by gravitational torques associated with
the orbital motion. In the case of AGN there is no natural repository of
angular momentum. One resolution of this difficulty is that the angular
momentum be removed by hydromagnetic torques that operate in a similar
manner to those just analysed in the case of the hole itself (e.g. Coroniti ,
1985). This may be responsible for the collimation of double radio sources.
When the accretion rate is large, it has been proposed that an optically
thick radiation-supported torus is produced in orbit around the black hole.
Equilibrium models of these tori have been described and used to model
quasars. In many respects they are similar to the original superstars of Hoyle
and Fowler. In particular they appear to be prone to dynamical instability
(Papaloizou and Pringle, 1984, 198 5 ; Goldreich, Goodman and Narayan,
1986). The most unstable modes are non-axisymmetric and global in
character and can be interpreted as edge waves coupled near to their co
rotation radii . Numerical simulations are essential to understand the non
linear development of these modes and in particular to determine if they can
form a basis for a model of the viscosity (Hawley, 1986).
Accretion of stars can also be important. A spherical star cluster around a
massive black hole will probably not be able to supply enough gas, via
stellar collisions and tidal stripping to fuel a powerful quasar, although
lower levels of activity may be supported in this way (e.g . Duncan and
Shapiro, 198 3 ; David , Durisen and Cohn, 1986). The rate of supply of gas
may however be enhanced if the potential is triaxial (e.g . Norman and Silk,
1983) or the black hole is in a binary (e.g . Roos, 198 1). An interesting
mechanism proposed by Carter and Luminet ( 1 982) is that the tidal shock a
star receives on passage by a black hole may detonate a nuclear explosion
and lead to a sudden release of energy. However, more detailed calculations
have concluded that this does not occur in practice (Bicknell and Gingold,
1986). Massive black holes may actually be built up out of stars in a dense
cluster. In an impressive series of relativistic stellar dynamical computations
Shapiro and Teukolsky ( 1985) have shown how a collapse may begin.
However, it remains to be seen if the collapse will continue and whether or
not a large black hole can be made in this manner.
Extensive reviews of accretion onto black holes are to be found in the
books by Shapiro and Teukolsky ( 1983) and Frank, King and Raine ( 1985)
as well as the articles by Pringle ( 1 98 1), Begelman , Blandford and Rees
( 1984), Begelman ( 1985) and especially Rees ( 1984) .
8.3 Observational evidence for black holes
8 .3 . 1 Active galactic nuclei
8 .3 . 1 . 1 Early history
Although the discovery of quasars in 1963 is widely cited as initiating the
modem era of astronomy, three parallel lines of investigation were all
leading up to this discovery and already posed the problems with which
astrophysicists were soon to be confronted . Together, and with the
considerable benefit of hindsight, these observations also provided strong
clues as to the solutions of these problems.
Firstly, Fath ( 1 908) observed the spiral galaxy NGC 1068 (of course
before it was known to be an external galaxy), and discovered unusual high
excitation lines. Subsequent observations by Slipher ( 19 17) demonstrated
that it had a redshift of 1 100 km s - 1 and that the lines were broader than
could be accounted for by rotation of the galaxy. Other galaxies like NGC
1068 showing bright condensed nuclei and broad, high excitation emission
lines were studied by Seyfert ( 1 943) and belong to a class which now bears
his name. Interest in these galaxies waned somewhat until the late 1950s
when Woltjer ( 1959) realised that Seyfert galaxies constituted a few per cent
of all spirals and that consequently they must live for a few per cent of the age
298 R. D. Blandford
of the universe, at least 108 yr - much longer than the time it would take the
emitting gas to escape from the nucleus. As the powers o_f nuclei of the
brighter Seyfert galaxies were known to approach those of typical galaxies
( - 4 x 10 4 3 erg s - 1 ), it was already apparent that nuclear activity required
considerable energy ( � 1059 erg per galaxy). Furthermore, the existence of
smaller scale activity in our Galaxy and the Andromeda galaxy strongly
hinted that most bright galaxies had active nuclei . Woltjer also deduced that
a blue photo-ionising spectrum would be necessary to excite the emitting gas
and suggested that it was produced by a cluster of hot stars, an idea that was
contradicted by the spectroscopic measurements of NGC 1068 by Burbidge,
Burbidge and Prendergast ( 1959) , which required a smaller mass-to-ultra
violet light ratio than a normal distribution of stars.
Secondly, many radio 'stars', were identified with external galaxies
following the demonstration by Baade in 195 1 (using an accurate radio
position determined by Smith) that the second most powerful radio source
in the sky, Cygnus A (Reber, 1 944) was identified with an unusual galaxy at
redshift z = 0.05 (Baade and Minkowski, 1954). The radio power was
estimated to be 10 44 erg s - 1 , larger than the optical luminosity of a bright
galaxy. What was equally remarkable was the discovery of high excitation
emission lines in the galaxy spectrum. These were interp reted as evidence for
a galaxy collision , an idea which had extraordinary longevity in view of the
large power involved. In fact the energetics were strained even further when
Shklovsky and Ginzburg proposed that the radio emission be produced by
the synchrotron process and Burbidge ( 1959) showed that a total energy of
3 x 1059 erg was required to account for the intensity of the radio emission.
Furthermore, if the ratio of relativistic protons to electrons in the source
were to be similar to that found in galactic cosmic rays the minimum energy
would have to increase by a factor ten to - 3 x 106 0 erg , the rest mass
equivalent of 2 x 106 M 0 , enough to power the radio source for a billion
years.
Strong evidence against the colliding galaxy model also came from the
interferometric results of Jennison and Das Gupta ( 1953) who showed that
Cygnus A in fact comprised two separate regions of emission that straddled
the optical galaxy instead of being concentrated on it . This pattern was to be
repeated in many more radio galaxies (Matthews, Maltby and Moffet,
1962) . Powerful radio galaxies gave the first indications of the enormous
energies involved in the most extreme examples of AGN .
The third line of enquiry, although limited to a single example, was to
provide the link between the first two . In 19 17, Curtis discovered that the
Ast rophysical black holes 299
elliptical galaxy M87 was unusual in that it had a narrow linear feature now
called a jet protruding from its nucleus. In 1956, Hiltner discovered linear
polarisation in the jet which enabled Burbidge ( 1956) to infer that the optical
emission was synchrotron radiation by relativistic electrons and that at least
105 6 erg of energy was required .
So, by the time of the announcement of the discovery of the quasar 3C273
(Schmidt, 1963 ; Hazard , MacKey and Shimmins, 1963), there was good
evidence fo r ongoing nuclear activity , the copious production of relativistic
electrons with aggregate energy far in excess of what might be expected from
normal stars. This was even more strongly true of 3C273 which had an
optical power of 104 6 erg s - 1 and even a longer jet than in M87 seen at
,....,
both radio and optical wavelengths. The association of quasars with these
milder phenomena was clearly understood (e.g . Burbidge, Burbidge and
Sandage, 1964).
8 .3 .1 .2 Ea r ly models of quasa rs
In addition , to these observational precursors of quasars, the theoretical
ideas that were to prove useful were also being developed. In fact the first
Texas conference on relativistic astrophysics (Robinson, Schild and
Schucking, 1964) was convened prior to the discovery of the quasars that
were to dominate its discussions, stimulated in part by the theoretical
challenges posed by the double radio sources (Maran and Cameron, 1964).
Hoyle and Fowler had previously proposed a model of AGN that involved a
'superstar' forming out of 10 6 M 0 of radiation-dominated gas. This
,....,
would radiate away its gravitational energy, at the Eddington limit,

contracting in the process and was thought to be capable of powering the
powerful radio galaxies.
Almost as soon as quasars were discovered , it was realised that unless the
redshifts had a non-cosmological origin, their large powers required that
matter be converted into energy with very large efficiency ; probably larger
than the maximum value "' 0.007 associated with nuclear reactions. (In fact
it is only comparatively recently that it has been possible to put this
argument on a rigorous basis.) The release of gravitational energy was
therefore implicated and several authors suggested that superstars might be
responsible. If so, they would have to be very compact with surface escape
speeds close to the speed of light. Unfortunately, as Feynman (unpublished
work), Iben ( 1964) and others quickly showed, such a star would be driven
unstable by post-Newtonian corrections to the equation of hydrostatic
equilibrium unless the gravitational potential (divided by c2) were less than
300 R. D. Blandford
the ratio of the gas pressure to the radiation pressure, already a small
number. Superstars could therefore not form efficient machines for
converting mass into energy ; neither did they seem particularly promising
for accelerating relativistic electrons. Nevertheless, this model did seem to
be on the right track, because Smith and Hoffleit ( 1963) soon discovered that
quasars were variable on timescales � 1 month implying that the energy was
produced within a region � 101 7 cm across. (Curiously, the discovery of
even more rapid variability in Seyfert galaxies, was not made until 1968 by
Pacholczyk and Weymann (1968).)
Various efforts to stabilise superstars using rotation, turbulence, nuclear
reactions and a 'C-field' proved unconvincing. Although it was shown that a
low entropy (i.e. thin) disk could exist with a binding energy up to 0.38 times
its mass (Bardeen and Wagoner, 1969) this structure was liable to serious
instabilities (Salpeter and Wagoner, 197 1 ; Salpeter, 197 1) and would
probably either fragment into stars or thicken to a high entropy disk
radiating a significant fraction of the Eddington limit of unspecified
structure and unknown stability properties.
Apart from the problem of not releasing much energy before collapsing
there was the separate difficulty of accelerating relativistic electrons to
produce non-thermal synchroton radiation. For this reason several authors
proposed that the active agent within a quasar be a magnetised spinning
superstar or disk , called a spinar or magnetoid (e.g. Sturrock , 196 5 ;
Ginzburg and Ozernoi , 1964). A simple estimate based on the virial theorem
suggested that a mass M would be permeated by a magnetic flux
- 10 3 0 (M/M 0 ) G cm 2 . For a mass - 108 M 0 and a size - 10 1 6 cm as
indicated by the observations this suggested field strengths up to 10 6 G which
would cause the disk to evolve rapidly dynamically (e.g. Woltjer, 197 1).
Again most of the energy would be released explosively near the end point of
the evolution just prior to the formation of a black hole. Several suggestions
were made that the double radio sources might be formed as a consequence
of such an explosion or outburst from a magnetised spinning superstar.
Now these models essentially identified the formation of a black hole with
the end of the activity. By contrast , Greenstein and Schmidt ( 1964) realised
that it was ' . . . important to know if continued energy and mass output from
such a collapsed object are possible.' A clear answer to this question was
provided by Zeldovich and Novikov ( 1964) and Salpeter ( 1964). These
authors pointed out that gas accreting onto a black hole could release its
energy reasonably efficiently. In the case of a Schwarzschild black hole, gas
should orbit the hole in circular orbits and so the efficiency of energy release
As trophysical black holes 301
should be given roughly by the binding energy of the most bound stable
circular orbit (i .e. 0.06) . They also realised that if gas were made freely
,..._,
available the accretion rate could build up to a limit setting in when the
pressure of the escaping radiation equaled the attractive pull of gravity.
Under either optically thick or optically thin conditions, this is given by
(8 . 1 2 )
It is known as the Eddington limit. Associated with this accretion rate is a

mass-independent timescale for increasing the mass of the hole,
X
M (TT "" 5
tEdd = -o- = 1 0 8 yr. (8 . 1 3)
M Edd 4n Gmpc
It now seems rather surpri sing that black hole models of AGN were not
developed more rapidly (although the subsequent discovery of the
microwave background in 1 965 and in 1 967 must have distracted attention
away from quasars) . Part of the explanation was that it was hard to
understand how the infall of gas could co-exist with outflow as apparently
observed in the inner parts of several galaxies, including our own . Another
reason was that it seems not to have been realised that radio sources,
quasars and Seyfert galaxies required there to be a more or less steady
release of energy over timescales � 10 8 yr. M ost models seem to have
envisaged impulsive energy release through a scaled up solar flare, black
hole formation or binary coalescence (e.g . Hoyle and Fowler, 1963a,b;
Burbidge, 1964) . This contradiction was particularly acute for the double
radio sources like Cygnus A which showed evidence for ongoing nuclear
activity through the high excitation nuclear emission lines and yet were
modelled by two anti-parallel 'jets' created in a single explosion (e.g. Le
Blanc and Wilson , 1969) . It is quite ironic that two of the staunchest
defenders of the steady state cosmology Hoyle and Burbidge, should have
believed in 'big bangs' in galactic nuclei .
The papers of Zeldovich, Novikov and Salpeter supplied the foundation
for most current thinking about the centre powerhouse in AGN. However,
they were not seriously developed until Lynden-Bell ( 1 969) derived and
applied a theory of thin accretion disks in orbit around massive
Schwarzschild black holes. He argued that magnetic stresses within the disk
would produce an effective viscosity which would cause angular momentum
to be driven outwards at the same time as mass flowed radially inwards,
releasing its gravitational binding energy in the process. Lynden-Bell
initially followed Salpeter, in assuming that the hole was non-rotating.
302 R. D. Blandford
However, as Bardeen ( 1970) was quick to point out, a black hole in an AGN
would probably be rapidly rotating, having been spun up by the gas from the
surrounding accretion disk. In this case the Kerr metric is appropriate and
stable circular orbits have a binding energy per unit mass that can be as large
as 0.42c 2 when !18 approaches its maximal value of 1/2m. (The efficiency of
energy release is still 0.2 for a hole rotating with angular velocity equal to
0.75 of its maximal value.) Spinning black holes can therefore provide
extremely efficient machines for converting the rest mass of accreted gas into
radiative energy. Accretion disk models had a further advantage. As the
magnetic field lines would be mostly frozen in to the differentially rotating
disk and would therefore have to keep undergoing magnetic reconnection ,
strong , inductive electric fields would be created in the (relatively small)
reconnection volumes and these would be able to accelerate relativistic
electrons. However, in this and subsequent papers (Lynden-Bell , 197 1 ;
Lynden-Bell and Rees, 1 97 1), accretion disks were still associated more with
dead quasars, i .e. local , low power AGN. The active, high power objects
were identified with massive, uncollapsed disks .
The idea that quasars in the prime of life might be powered by accretion
disks around black holes gradually gained ground matching contemporary
developments in the study of binary X-ray sources (e.g. Rees, 1977). At the
same time, the principal competition , models involving dense clusters of
stars, seemed to lose ground as the observational position developed.
8 .3 . 1 .3 Star clusters
A competitive class of quasar models to those involving single coherent
objects developed in parallel to models involving massive black holes. This
involved dense star clusters (e.g. Gold, Axford and Ray, 1 964; Spitzer and
Saslaw, 1 966 ; Colgate, 1967 ; Sanders, 1970). In initial versions of this model
it was proposed that quasars were the consequence of an exaggerated rate of
supernovae within the cluster. 108 M 0 of high mass stars in a volume
1 pc3 can qualitatively reproduce many of the observed features of a
,...,,
quasar, the ionising radiation coming mostly from the hot stars , the
outbursts being attributed to the supernovae themselves and the non
thermal emission deriving from the overlapping supernova remnants. In a
star cluster this dense, the relative velocity of the stars becomes comparable
with their surface escape speeds. This implies that physical collisions become
competitive � ith gravitational collisions. Unfortunately, the consequences
of these collisions have never properly been understood. If the stars coalesce,
then rapidly evolving high mass stars can be assembled . It was even
suggested that supernovae might trigger one another in a sort of chain

reaction. A cluster of this type would inevitably produce a lot of interstellar
gas which may create additional stars and accelerate the dynamical
evolution of the cluster (Bisnovatyi-Kogan and Sunyaev, 1972; Begelman
and Rees, 1978).
The principal objection to star clusters was that, at their best, they were
never very efficient. Typically, � 10 - 3 of the mass of the stars can be
converted into radiation over the life of the cluster. This difficulty may be
ameliorated by replacing the stars with compact objects - black holes,
neutron stars or possibly white dwarfs (e.g. Gunn and Ostriker, 1970;
Arons, Kulsrud and Ostriker, 1975; Stoeger and Pacholczyk, 1983). A newly
formed pulsar can store up to 1 per cent of its rest mass as spin energy
(although observations of galactic pulsars argue against this actually
occurring : Lyne, Manchester and Taylor, 1985). Stellar mass black holes
accreting in a dense gaseous environment should be able to radiate with a
similar efficiency per unit mass of accreted gas to that invoked for stellar
mass black holes (i .e. 10 per cent).
"'
However, a second difficulty is harder to overcome. This is that large

outbursts in powerful quasars radiate much more than 105 2 erg , the
maximum energy that can be derived from a single star. Attempts to explain
this in terms of non-linear or collective interactions have been less than
convmcmg .
A third and more direct problem with star cluster models arose when it
became clear that most double radio sources had central components which
could be resolved using VLBI. This realisation, in conjunction with the
production of high resolution maps of the outer radio components which
contained b right 'hot spots' , encouraged Rees ( 197 1), Longair, Ryle and
Scheuer ( 1973), and Scheuer ( 1974) to develop models in which the radio
components are fueled continuously through channels or jets. Blandford
and Rees ( 1974) suggested a specific collimation mechanism for the jets - the
formation of a pair of De Laval nozzles along the spin axis of a dense
rotating cloud of gas. This model did not specify the nature or origin of the
jet fluid , only requiring that it be relatively hot and light. Unfortunately, in a
powerful source like Cygnus A, the nozzles must form on a scale 100 pc, "'
far larger than the size of the compact radio sources which are typically
"' 1 pc in size. This implies that the collimation of the jets occurs on scales
� 1 pc. There was no straightforward way to accomplish this in the context
of a massive star cluster, especially when it was discovered that the source
axes on the VLBI scale were roughly parallel to the larger scale structure.
304 R. D. Blandford
Attempts to concentrate a star cluster into a volume � 1 pc3 only hasten its
evolution into a massive black hole (Begelman and Rees, 1978 ; Shapiro and
Teukolsky, 1985). A stable compact gyroscope, i.e. a single spinning object,
rather than an assemblage of small, independent radiators was suggested.
8.3 . 1 .4 Recent evidence for black holes in AGN

Further observational developments have strengthened the belief that AGN
contain massive black holes. X-ray ·variability appears to be common in
Seyfert galaxies, and in a minority of lower power examples, this has been
seen to occur on timescales as short as minutes (e.g. Tennant et al., 198 1 ;
Marshall et al., 1983). I n fact Barr and Mushotzky ( 1986) have claimed that
the X-ray variability timescale is correlated with the X-ray luminosity in the
form (t/m) "' lOO(Lx /L Edd ) which implies that the X-rays originate from
within roughly a hundred Schwarzschild radii. (Even more rapid variability
on timescales 30 s has been reported from the BL Lac object H0323 + 02
,...,
(Feigelson et al., 1986), although here it is suspected that the variability

timescale may be unrepresentatively short compared with the size of the
emitting region on account of relativistic beaming.)
The stronger hypothesis, - that essentially all bright galaxies harbour one
or more black holes in their nuclei - received a boost when it was found that
low level emission line and radio activity could be detected in the majority of
local galaxies (e.g. Keel , 1985). More recently, Filippenko and Sargent
( 1985) discovered that broad (i.e. high velocity) wings to the emission lines
are commonplace and have therefore argued that low-level , quasar-like
activity must generally be present in galaxies.
Superluminal expansion of radio components, first discovered in 197 1
(Moffet et al., 197 1 ; Cohen et al. , 197 1 ; Whitney et al., 197 1 ) has been shown
to be a fairly common property of compact radio sources (e.g. Cohen, 1986).
This phenomenon, in which two or more distinct radio components
separate with apparent speeds in excess of c, was actually predicted by Rees
( 1966). It is generally interpreted as an illusion caused when radio-emitting
plasma moves with relativistic speed at a small angle to the observer's line of
sight. Although the kinematic details remain an area of active controversy,
the evidence that there is outflow at relativistic speed is strong and this, in
turn, is circumstantial evidence for the presence of a relativistically deep
gravitational potential well, i.e. a black hole.
Recent work on faint quasar counts (Koo , 1986; Boyle et al. , 1987) has
allowed an improved estimate of the overall efficiency of quasars (Soltan,
1982; Phinney, 1983). The number of quasars with optical flux in excess of
some value F rises with decreasing flux oc F - 2 · 3 down to a flux equivalent to

19m . The number of fainter quasars rises far less rapidly so that the mean
energy density of quasar light observed at earth converges at 22m where ,......,
there are N Q ,......, 4 x 106 quasars on the sky. An average magnitude for these
quasars of 20. 5m corresponds to a flux F Q "" 2 x 10 - 1 3 erg cm - 2 s - 1 if each
quasar radiates like an AO star. However, quasars must radiate significantly
more UV radiation than an AO star in order to account for the observed
emission lines, and so we must multiply this flux by a bolometric correction
which we estimate as fuv ,......, 3 . The mean redshift of these quasars is < z) 2 ,......,
and so each quasar photon will have its energy reduced by a factor
,......, ( 1 + (z) ) - 1 as the universe expands. The mean energy density of quasar
light observed at earth is then U Q ,......, N 0 F 0 fu v O + (z) )/c ,......, 1 . 5 x
10 - 1 6 erg cm - 3 - 3000 M 0 Mpc - 3 . (This is a conservative estimate. Any
increase in the number of quasars or their bolometric corrections would
only increase the required efficiency.) Now, the local density of bright
galaxies is roughly ,......, 10 - 2 h - 3 Mpc - 3 , where h is the Hubble constant in
units of 100 km s - l Mpc - 1 . Therefore, at least 3 x 105 M 0 of energy
equivalent must be released per bright galaxy. Now if we introduce an
efficiency of conversion of rest mass into radiant energy s ,......, 0. 1 , the remnant
nuclear masses per bright galaxy must be ,......, 3 x l06(s/O. l ) - 1 h - 3 M 0 , if we
assume that quasars .are powered gravitationally:
As we discuss below, it appears that most bright galaxies have central
masses < 3 x 10 6 M 0 . We can also probably already exclude the possibility
that 10 per cent of nuclear remnants are heavier than 3 x 1 09 M 0 . Hence,
the required efficiency of energy production must exceed ,......, 10 - 3 (h = 1), or
,......, 10 - 2 (h = 0.5). This virtually rules out star cluster models, as well as black
hole models that postulate super-critical accretion with low radiative
efficiency (e.g. Abramowitz, Jaroszynski and Sikora, 1978).
As we discuss below, the strongest evidence for black holes in X-ray
binaries is dynamical . Unfortunately, unlike in the stellar mass case, the
alternatives to black holes (e.g. spinars, magnetoids etc.) are far less well
specified than neutron stars, and therefore cannot be so easily ruled out on
the basis of mass determinations. There have been several attempted
dynamical determinations of the central masses of AGN .
The observational study of the nucleus of M87 by Young et al. ( 1978) and
Sargent et al. ( 1978) showed that there was a central cusp of light over and
above the central core of stars and that the measured velocity dispersion of
the stars rise� within the central ,......, 100 pc (cf also de Vaucouleurs and
Nieto , 1979). They interpreted these observations in terms of a model
306 R. D. Blandford
involving a central black hole of mass ....., 3 x 109- 5 x 109 M 0 . (This

conclusion has been challenged by Dressler, 1980.) It is known that not all
bright galaxies show evidence for central cusps (Young et al. , 1978). If we
take this evidence at face value, we find that we are only probing the stellar
dynamics on a length scale equivalent to 3 x 105 Schwarzschild radii. There
has therefore been some attention paid to dynamical explanations of these
observations not involving a black hole. One possibility is that the central
star cluster has an anisotropic velocity dispersion so that the preponderance
of radial orbits near the centre is able to reproduce the observed cusp
(Duncan and Wheeler, 1980). Newton and Binney ( 1984) and Richstone and
Tremaine ( 1985) have shown that the observations can be consistently
interpreted using a constant mass to light ratio (M/L B ....., 10 and a "" 2 : 1
radial velocity anisotropy. However, i t i s not clear that an equilibrium
stellar distribution of this type is stable against velocity space instabilities
(e.g. Barnes, Goodman and Hut, 1986). Related observational work has
been carried out by Tonry ( 1 984) , Dressler ( 1 984) , Davies and Illingworth
( 1986) and especially Kormendy ( 1 986) on the nuclei of the galaxies M 3 1 ,
M32, and NGC 1052. Rises in both the central velocity dispersions and the
rotational velocity are reported, consistent with central black hole masses
....., 106 - 107 M.0 . However, the worry about anisotropic velocity dispersions
remains, and we can strictly only regard these masses as upper limits at this
stage.
Further details on these and related topics are to be found in the
proceedings of four recent conferences edited by Dyson ( 1985), Miller
( 1 985), Giuricin et al. ( 1986) , Swarup and Kapahi ( 1986) .
8.3 . 1 .5 Future observations

It is of interest to speculate upon what future observations might give firmer
dynamical evidence for massive black holes in AGN . The most immediate
hope lies with Space Telescope now, regrettably, scheduled for launch no
earlier than 1988. Its linear resolution should improve on that of ground
based telescopes by at least a factor 10 and so it should be able to resolve the
stellar cusps of galaxies quite well (if they are present). In particular, if the
central velocity dispersions should rise by a further factor "" 10 1 12 "" 3 , this
will probably rule out anisotropic velocity dispersion models.
More speculatively, stronger evidence for massive black holes would be
forthcoming if a single example of a star or another black hole in orbit could
be found. As discussed by Rees ( 1982) in the context of our Galactic centre, a
single star orbiting outside its tidal radius close to an underluminous black
hole might modulate the light output with a regular period � 1 h. The effects
of Lense-Thirring precession might even be detectable. Searches for low
power periodic signals in the continuum emission of selected objects might
turn up an example of this.
Alternatively, when two galaxies (each possessing black holes in their
nuclei) merge, the hole associated with the smaller galaxy will be dragged
into the nucleus of the larger galaxy in a few orbital periods. Thereafter, the
dynamical evolution time will increase to ""' 109 yr as the interaction with
individual stars takes over (Begelman , Blandford and Rees, 1980 ; Roos,
198 1). The holes will then spiral together until they become so close that
gravitational radiation by the black hole binary takes over. Now, if we
assume that the heavier black hole has a mass m 1 = 10 6 m 1 6 M 0 , and the
lighter one a mass m 2 = 10 6 m 2 6 M 0 , then the lifetime to coalescence will
be related to the orbital period P yr by T,..., 4 x 1 07P 8 1 3 m:;_l1 3 yr. Searches
for regular peri odic behaviour in quasars (e.g . Ozernoi and Chertoprid,
1969) have not yet been fruitful and we would not expect them to contain
binaries with measurable periods. However, it is just possible that a few
lower power Seyferts may harbour binary black holes with detectable
periods � 1 yr. A black hole binary would probably perturb the central
stellar distribution so strongly that the black hole would be supplied �ith
gas and the nucleus would be active at this time. For black hole masses
m 1 6 ""' m 2 6 � 1 , and an assumed merger rate of one per galaxy, roughly 1 in
300 galaxies could contain binaries and show periodic behaviour with
P � 1 yr. (As we discuss below, Lacey and Ostriker ( 197 5) propose that
binaries of this type occur very frequently in galactic n1=1clei.) These binaries
would probably not be seen as spectroscopic binaries because the sizes of the
broad emission line regions are inferred to be significantly larger than the
radii of these hypothetical binary orbits. However, they may be detectable as
eclipsing binaries if both black holes are accompanied by extensive accretion
disks. A regular monitoring program of nearby Seyfert and LINER nuclei
might possibly uncover an example.
8 .3 . 1 .6 Possible observations incompatible with the presence of massive

black holes
In the same spirit, it is worth discussing some conceivable future
observational discoveries which might be capable of ruling out the presence
of massive black holes in AGN . One of these is the detection of large
amplitude, high Q pulsations with timescales of order a day or larger in a
powerful quasar. This could probably not be attributable to modulation of
308 R. D. Blandford
the emission from an accretion disk by the orbit of a single star as discussed
above and would instead probably represent the presence of a massive
uncollapsed object . Another possibility is the detection of X-ray or y-ray
lines with a common redshift relative to the host galaxy of z - 0. 1--0.2. This
would certainly revive inter� st in neutron star cluster models of AGN.
Alternatively, Space Telescope (operating with its advertised resolving
power) may fail to detect any evidence for dormant central masses in the
nuclei of local galaxies. As the arguments above imply, this would call into
question far more than the black hole model (especially if h - 0.5).
However, perhaps the most telling argument against black hole models
would come if it could be demonstrated that the compact optical and X-ray
sources are generally displaced from the dynamical centres of their host
galaxies as well as each other. An argument similar to this has already been
used to limit the masses of black holes perhaps associated with X-ray
sources in globular clusters. More relevantly, it has been applied to our
Galactic centre and it is illustrative to discuss the current controversy as to
whether or not our Galactic centre contains a massive black hole.
8 .3 . 1 . 7 Our Galactic centre

The Galactic centre region, most recently reviewed by Brown and Liszt
(1984) and Lo ( 1986) contains a compact radio source with unique spectral
properties known as Sgr A* (Balick and Brown, 1974) . This source is more
compact than - 3 x 101 4 cm (Lo et al. , 198 5) , and is surrounded by a ring of
gas and dust of radius - 2 pc apparently connected to Sgr A* by three
curving radio features. These radio 'arms' can also be observed in 12.8 µ
infra red Ne II lines and although the kinematical configuration is still
unclear, a central mass of - 3 x 106 M 0 is strongly implied (Crawford et al. ,
1985). A n even higher velocity dispersion feature was discovered by Hall,
Kleinmann and Scoville ( 1982) in He I lines originating from the vicinity of
Sgr A*. Naive application of the virial theorem gives a similar central mass
although this feature is also interpretable as an outflowing wind . In
addition , Sgr A* is located near the centre of a cusp of stars, traced out by
their infra-red emission. It therefore might seem natural to assume that a
- 3 x 106 M 0 black hole is present in Sgr A*.
However, there i s also an unusual infra-red source located nearby known
as IRS 16. This source which, is relatively blue, is argued to be the source of
the photo-ionising flux required within the central pc and to comprise - 100
hot (BO) stars. However, Sanders and Allen ( 1986 and references therein)
have shown that IRS 16 is comprised of several subsources and is displaced
from Sgr A * by - 10 1 7 cm. So, we must choose between I RS 16 and Sgr A*

as the centre of the galaxy . There are already some clues. We . know on
dynamical grounds that any million solar mass black hole must settle very
close to the centre of the star cluster. If the black hole is identified with
Sgr A*, then its tidal field would destroy a small star cluster like I RS 16 in a
timescale "' 1000 yr, far shorter than the age of the stars. On these grounds,
Allen and Sanders argue that the mass in Sgr A* must be much less than
3 x 10 6 M 0 , typically - 100 M 0 . However, VLBI observations (Backer and
Sramek, 1987 and private communication) have already shown that Sgr A*
is at rest (to less than - 50 km s - 1 ) with respect to the dynamical centre of
the Galaxy. This measurement would already seem to rule out the
possibility that Sgr A* is in orbit about the Galactic centre (and also that it is
an outflowing radio jet). This implies that Sgr A* and not I RS 16 is the
Galactic centre. It seems that independent of the black hole hypothesis, we
must consider that either there is an unlikely superposition of unusual
objects or that I RS 16 is a transient object. A final choice will probably be
made on the basis of an improved limit on the proper motion and infra-red
radial velocity mapping.
The Galactic centre is also a source of variable y-rays, notably the 5 1 1 keV
positron annihilation line. If we want to associate this emission with a
relativistic electron-positron wind (with a discharge N � 10 4 3 s - 1 ) flowing
out from the Galactic centre, then we would naively,expect that the size of
the source of the wind l would be determined by the condition that there is
an optical depth of order unity to photon-photon pair production in the
source, i .e. l - NaT /4nc - 3 x 107 cm . Using this argument, Lingenfelter and
Ramaty ( 1 982) have suggested that the positron source is a - 100- 1000 M 0
black hole for which l is comparable to the Schwarzschild radius and for
which the total luminosity is similar to the Eddington limit.
Another difficulty with postulating the presence of a massive black hole in
the Galactic centre has been emphasised by Gurzadyan and Ozernoi ( 198 1
and references therein) . This is that if we use the standard estimate of the rate
of swallowing stars by a black hole (e.g. Lightman and Shapiro, 1977), then
the galactic centre would be far more luminous than it is observed to be and
the black hole would have grown to a far larger mass by now. However, Rees
( 1982) has argued that stars will only be captured occasionally, and that
when they are, the stellar debris is mostly ejected very quickly by the energy
liberated as a small fraction of the star's mass accretes on the hole. The mass of
a black hole in a galaxy like ours could therefore stabilise at 106 M 0 and
-
310 R . D . Blandford
the fraction of the time when it is emitting at a r<:lte comparable with the
Eddington limit will be correspondingly small.
8 .3 .2 Black holes in binary X-ray sources

8 .3 .2 .1 Early history
As we described in Section 8 . 1 , the idea that black holes are a natural end
point of stellar evolution of massive stars is over fifty years old and,
compared to the situation with AGN , there has been comparatively little
resistance to claims that they exist in binary stars. However, the discovery of
Seo X- 1 in 1962 (Giacconni et al., 1962) and its identification with a 1 3m star
were not immediately heralded as discoveries of neutron stars (Shklovsky,
1967) or black holes despite the fact that several physicists (e.g. Chiu ,
Salpeter, Wheeler) were getting interested in their theoretical properties.
Neutron stars were (rightly) not regarded as promising prospects for optical
observation because of their small ( - 101 3 cm) surface areas. It was
however, realised that newly formed neutron stars should be hot with
temperatures ( - 107 K) and should be observable for a cooling time - 1000
years as X-ray sources. Indeed it was explicitly suggested (e.g . Chiu , 1964)
that the few discrete X-ray sources known at the time might be neutron stars
made in recent supernovae.
Again, it was Zeldovich ( 1964a) who appears to have been the first to
realise that a compact object in a mass transfer binary or just conceivably a
dense interstellar cloud would be a more prodigious source of X-rays. The
potential energy released when a gram of gas falls onto a neutron star or
black hole is - 10 2 0 erg , far larger than the mean amount of energy per gram
of neutron star that can be transported by degenerate electron conduction to
the surface before neutrino emission cools the interior (i .e. - 10 1 2 erg). He
also noted that this emission mode was available to black holes despite the
absence of a hard surface, because gas would probably not approach the
event horizon smoothly but , instead , separate gas streams could collide with
one another, shock, heat and then radiate away their internal energy.
Accretion with angular momentum is likely to be more radiatively efficient
than spherical accretion (Zeldovich, 1964b) . It was also realised that the best
evidence for the presence of a black hole was likely to be dynamical and
several binary star catalogues were searched for massive unseen
comparisons (e.g. Zeldovich and Guseynov, 1965; Trimble and Thome,
1969) of which several were found . (Unfortunately, none of these candidates
has survived .)
Astrophysical black holes 31 1
8.3.2.2 Cyg X-1

The observational position improved dramatically with the launch of the
Uhuru satellite in 197 1 . One of the first sou rces to be investigated was
Cygnus X- 1 . It was a powerful X-ray source but, unlike several other
powerful sources, it did not exhibit regular pulsation (Schreier et al., 197 1 ).
Instead Cygnus X- 1 had an unusually hard X-ray spectrum and displayed
rapid flickering on the timescale of a few milliseconds. The optical
identification was with a bright blue supergiant HD 226868 of spectral type
09.7Iab . Optical spectroscopy by Webster and Murdin ( 1972) and Bolton
( 1972) showed spectral lines whose Doppler shifts varied sinusoidally with a
period Pb = 5.6 d and a velocity amplitude K = 76 km s - 1 coming from the
surface of the supergiant. (They also showed features that appear to
originate from gas streaming toward the X-ray emitting companion.) Now if
mx is the mass of X-ray emitting star and m0 is the mass of the supergiant, we
can deduce the value of mass function
m x3 sin 3 i Pb K 3
( mx + m0 ) 2 2 n G
f(m0, mx, i) = = - 0.25 M 0 ,
where i is the angle of inclination between the orbital angular momentum

vector and the line of sight.
Now, given the mass function we see that mx increases monotomically
from zero as m0 and cos i increase from zero. However, a normal blue
supergiant of this spectral class would have a mass ,..,,, 30 M 0 . Furthermore,
the X-ray emission also exhibits the 5.6 d period, confirming the
identification. However, eclipses are not seen in the high energy X-rays.
(Partial eclipse is seen at low energy which is attributed to photoelectric
absorption in the outflowing wind from the supergiant.) The inclination
angle is then probably less than 60°. Together these values imply that
mx � 8 M 0 comfortably larger than the Oppenheimer-VolkotT mass for a
neutron star. Not surprisingly, Webster, Murdin and Bolton concluded that
it was 'inevitable that we speculate' that the comparison to HD 226868 was a
black hole.
Unfortunately, this argument is contestable as it stands. The main
difficulty is that the supergiant may well have lost a lot of mass over the
lifetime of the system (not necessarily to the X-ray source but possibly out of
the binary altogether). It may therefore have a very much smaller mass than
its spectral type would conventionally indicate and perhaps be a hot
subdwarf (Trimble e t al., 1973). For this reason, Paczynski ( 1974) showed
that it was possible to place a lower bound on mx independent of the
312 R. D. Blandford
evolutionary history of its optical companion. To do this, we note that

observations of the spectrum of the optical star allow one to deduce its
effective temperature and its surface gravity. What we cannot measure is the
distance d to the binary. Now the radius of the star will be R0 oc d and the mass
will scale m0 oc R 1 1 2 oc d1 1 2. Now, as we know the binary period, we can
rewrite the mass function as mx sin i oc a2 where a is the separation of the
stars. However, the absence of eclipses implies a > R0 sec i oc d sec i. Hence,
by minimising cosec i sec 2 i with respect to the unknown inclination i we can
put a distance-dependent lower bound on mx , i .e.
mx � 3.4 M 0(d/2 kpc)2. (8. 14)
Now detailed studies of the reddening of HD 226868 and nearby stars
allow one to place a lower bound on d (Margon et al. , 1973 ; Bregman et al.,
1973), which indicates a distance in excess of 2 kpc which, in turn implies that
mx > 3 .4 M 0 . Of course local patchiness in the distribution may allow Cyg
X- 1 to be closer than 2 kpc, but already this is looking rather contrived . In
fact the system can be further constrained by modelling the ellipsoidal light
variations associated with the tidal deformation of the optical star and the
rotational broadening of the spectral lines, and the results indicate that the
star is indeed a high mass supergiant as indicated by its spectral type (Avni
and Bahcall, 1975). The most recent spectroscopic study of this system, by
Gies and Bolton (1986) concludes that the minimal (probable) masses are
mx = 1 M 0 ( 16 M 0), m0 = 20 M 0 (33 M 0). More contrived models
involving triple systems and X-ray generation by normal stars have been
proposed, but the observations show none of the expected features (e.g.
Bahcall, 1978).
In conclusion, the case that Cyg X- 1 has a mass in excess of the
Oppenheimer-Volkov limit is, by astronomical standards pretty strong
and, importantly, has strengthened significantly since 1972. It seems that the
case for calling it a black hole is as strong as the Oppenheimer-Volkofflimit
which is computed to be - 1 .6-2 M0 for reasonable equations of state
(Arnett and Bowers, 1977) and - 3 M0 for equations of state constrained
solely by the requirement that general relativity be the correct theory of
gravity (e.g. Rhoades and Ruffini , 1974).
8.3.2.3 LMC X-3

More recently, LMC X-3, an X-ray source in the Magellanic clouds has been
associated with a young, hot star on the basis of an extremely good X-ray
position. The optical star has a binary period Pb = 1 .7 d and K = 235 km s - 1
(Cowley et al. , 1983). This source has two favourable features. Firstly, the
mass function is computed to be 2.3 + 0.3 M 0 and secondly the distance to
the Magellanic clouds is known fairly accurately to be d = 55 kpc. Repeating
the arguments listed above for Cyg X- 1 , we find that the B3V spectral type
optical companion should have a mass "'"' 7 M 0 if there has been no unusual
evolution. Independent of the evolutionary history, the avoidance of X-ray
eclipses requires that mx � 10 + 4 M 0 (Paczynski, 1984).
Unfortunately, there is a possible complication in this system because the
accretion disk around the X-ray source may contribute a significant portion
of the optical emission from this binary allowing both the compact object
and its companion to be of lower mass. In particular, the mass of the X-ray
source may possibly be as low as -- 2.5 M 0 (Mazeh et al. , 1986),
uncomfortably close to the Oppenheimer-Volkoff limit. In this case the
companion would have to be a � 1 M 0 helium star, a possibility that should
eventually be ruled out using high dispersion spectroscopy.
8 .3 .2 .4 A0620-00
Most recently, McClintock and Remillard (1986) have shown that the
orbital period for an X-ray nova A0620-00 is a spectroscopic binary. (An X
ray nova is a transient X-ray source that becomes very bright for several
months and then fades away.) Fortunately, A0620-00 also brightened at
optical wavelengths and so there is no doubt about the identity of the
companion. McClintock and Remillard found thaLPb = 7 .75 h and K =
457 km s - 1 giving a mass function f(m0 , mx , i) = 3 .2 + 0.2 M 0 . Again, there
are no X-ray eclipses and from the spectral type of the optical companion
(K5V) and modelling of the light curve, it is deduced that mx > 3.2 M0
(McClintock, 1986). This system is unusual in one other respect. During its
1975 outburst, it became the brightest X-ray source in the sky. However,
during quiescence it is unobservably faint in X-rays and so appears to
radiate far less in X-rays than would be expected on the basis of the optical
emission from the accretion disk. Somehow or other the X-ray energy that
should be released by gas as it spirals in towards the black hole is being
absorbed or scattered out of the line of sight. (Alternatively, it is possible
that the gas accumulates in a reservoir in the outer parts of the disk.) Until
the reason for this behaviour is understood, there must be some small
residual doubt about the inferred geometry of the binary.
8.3 .2.5 LMC X-1
A fourth source, LMC X- 1 , has been put forward as a black hole candidate
on dynamical grounds by Hutchings et a l . ( 1983). The optical identification
314 R. D. Blandford
of this source, which is also found in the Large Magellanic Cloud is

less certain. The companion is believed to be ·a bright 07 III-V star.
Recent discovery of a soft X-ray photo-ionised nebula centered on this star
(Pakull and Angebault, 1986) provides confirmatory evidence that the
identification is correct because the high ionisation states observed in the
nebula could not be produced by a normal star. Unfortunately no X-ray
modulation at the optical period of Pb = 3.9 d has been seen. Using K =
66 km s - 1 , a mass function f(m0, mx, i) = 0 . 12 M 0 is obtained, although

there is still some uncertainty about this as other periods Pb ::::;; 1 d may fit the
data as well. The absence of X-ray eclipses can then be used to argue that
i ::::;; 64° . In addition, the requirement that the bolometric luminosity of the
optical star be less than the Eddington limit leads to a lower bound on the
companion mass m0 � 8 M 0 with a probable black hole mx "' 4 M 0 . Future
spectroscopic observations should either rule out or strengthen the black
hole candidacy of this system.
8.3.2 .6 SS433
An extensively studied galactic object which has also been argued to contain
a black hole· is SS433. SS433 originally appeared in a list of emission line
stars. After some false trails (chronicled in Margon, 1980), Clark and
Murdin (1978) realised that it was coincident with an X-ray source and a
compact radio source at the centre of a large supernova remnant W50 which
they naturally presumed were related. They also noted that, in addition to
the unusually strong and broad hydrogen and helium emi ssion lines, several
weaker unidentifiable lines were also present in the spectrum. In a more
careful study of the spectrum, Margon et al. (1979) identified the weaker
lines with redshifted and blueshifted hydrogen and helium transitions.
Furthermore, the Doppler shifts changed from night to night in a systematic
manner.
Among a host of ephemeral theoretical suggestions, was one by Fabian
and Rees (1979) who suggested that the emission be produced by two
oppositely directly jets, analogous to those found in the double radio
sources, but in this case containing dense clouds of emitting gas moving with
relativistic speed . Milgrom (1979) further speculated that the modulation of
the Doppler shift may be due to precession of these jets about an axis fixed in
space. Further observations by Abell and Margon (1979) verified this
kinematic model . Using several periods of data, we now know that the jet
speed is stable at v =0.26c , the vertex angle of the cone on which the jet
precesses is e = 20° the inclination of the cone axis to the line of sight is
'
j = 80° and the precession period is Pp = 163 d (Margon, 1984).
SS433 also exhibits a 1 3 . 1 d, 0.7m photometric double-peaked variation
(as well as a slower 163 d variation). This is attributed to mutual eclipsing of
an accretion disk surrounding a compact object and its binary companion
orbiting with a 1 3 . 1 d period (Leibowicz, 1984). This same period is also
detectable in small radial velocity variation of the brighter, 'stationary'
emission lines. However, the hydrogen lines vary with a smaller velocity
amplitude and different orbital phase to the higher excitation He II lines. It
is important to determine which of these two radial velocity variations
should be used to determine the mass function, because the former
possibility gives f(m0, m x , i) - 0.5 M 0 and a mass m x - 1 M 0 for the
compact object (which could presumably be a neutron star) and the latter
suggests f(m0 , mx , i) - 1 1 M0 implying a black hole companion. Now,
Crampton and Hutchings ( 198 1) have argued that the higher excitation
helium lines should be produced closer to the photo-emission source,
presumably the compact object, and that the hydrogen lines are associated
with gas streams hitting the outer edge of the accretion disk in a hot spot.
Furthermore, the phase of maximum He II radial velocity is roughly 90°
different from the primary minimum in the photometric variation which is
entirely consistent with the gas stream interpretation.
Further confirmatory evidence that the He II lines measure the velocity of
the compact object and consequently that SS433 is a high mass system is
provided by the discovery of orbital modulation of the 'moving' emission
lines. The wavelength A0 of a spectral line emitted by gas in either jet is
related to the emitted wavelength Ae by
A0 = 2 Ae 2 1 2 [c + v (sin j sin 8 cos t/f + cos j cos 8)] , (8 . 15)
(c - v ) 1
where the + signs refer to the two jets and t/J is the precessional phase which
can be written at 2nt/PP + t/f 0 in the absence of orbital modulation. Now, on
quite general grounds, we expect the star in orbit about the compact object
to nutate and to perturb the disk (together with the jets which are generally
assumed to be launched perpendicular to the disk) at half the synodic
period, Ps = 1/2(Pb- 1 + P; 1 ). If, as we shall find, the precession is retrograde
with respect to the orbital motion, we must choose the plus sign and Ps =
6.06 d. (Retrograde precession will result if the compact object exerts a
torque on the star and the disk is slave to the star, or the star forces the disk
to precess directly. Lense-Thirring precession would be prograde and is
3 16 R. D. Blandford
therefore ruled out.) In general, we might expect j, a and I/I to contain

contributions oc cos(2n t/P + const.) which would create sidebands in a
5
frequency analysis of the measured Doppler factors. Now in addition to the

fundamental precessional period of 163 d, two additional periods are found
at 6.28 d and 5.83 d = (P5- 1 + P; 1 ) - 1 . No signal is detected at Ps 6.06 d. If
=
we now inspect equation (S . 15), we find that these sidebands can be

reproduced if the precessional place I/I and the cone angle () are modulated
by the orbital motion. This confirms that the precession is retrograde (Katz
et al. , 1982 ; Newsom and Collins, 1982). Now, given this kinematical
interpretation of the sidebands in the Doppler shift data, we can compare
the phase of the modulation with the orbital phase and again we find that it
is consistent with the phase of the He II lines (Margon, 1984).
We therefore have a fairly comprehensive kinematical description of the
SS433 systems (Fig. 8.7). What is quite disturbing is the absence of
convincing dynamical explanations for the jet collimation, unique speed,
precession and orbital modulation. (For a critique of various possibilities,
see Margon, 1984; Collins, 1985.) Nevertheless, if we take the kinematics at
face value, and assume that the jet precesses about the orbital angular
momentum . (i.e. i j) then the mass function implies that m�
= =
l l(m0 + mx) 2 M 0 . The final stage in the argument is to put a lower bound on
the mass of the companion star m0 � 20 M0 from the requirement that it fill
its Roche lobe and be able to radiate the observed luminosity of - 8 x
10 3 8 erg s - 1 (quoted by Band and Grindlay, 1985). The mass of the compact
object is then mx � 7 M 0 , indicating a black hole and consistent with it
radiating - 10 3 9 erg s - 1 at the Eddington limit.
However, as was the case with Cyg X- 1 , the compact object could itself be
a close binary with orbital period Pb 2 - 1 . 5 d and precessional period Pp "'
4 P;/3 Pb 2 cos fJ. A recent study of this possibility by Fabian et al. (1986)
shows that it is possible to devise evolutionary pathways to this
configuration. This model allows the compact object to be a neutron star.
However, it seems hard for this system to avoid producing a further
modulation of the optical emission with a period - Pb i · So far, this period
has not been detected.
In conclusion, the weight of evidence favours SS433 containing a black
hole. However, the arguments for this are less direct than in Cyg X- 1 . The
importance of SS433 is that it shares some properties with active galactic
nuclei (relativistic precessing jets, winds, broad emission lines etc.) and it
may be a miniature version of a quasar.
8.3 .2 . 7 Other black hole candidates

We have described the evidence for black hole candidacy for fiye X-ray
sources in decreasing order of reliability. It is surprising that more examples
have not been found since 1972. We have seen that the identification, and
orbit must both be secure and we must usually add an astrophysical
argument to limit the range of inclination angle before we can assemble a
strong case for having a black hole binary. In the past, various other criteria
have been proposed for black holes but these have not turned out to be
reliable. Nevertheless it is instructive to discuss why these candidates are no
longer taken seriously.
Firstly, the X-ray quiet binary stars that have massive, apparently
invisible companions such as e Aur (e.g. Cameron, 197 1), f3 Lyr (e.g.
Fig. 8.7 . Kinematic model of SS433. Two antiparallel jets emanating from a
compact object in a binary system precess on a cone with opening angle
0 - 20° and period PP - 163 d . Tidal torques produced by the orbiting
companion impose a small nutation on the jets with a 6 .06 d period .
(Adapted from Collins, 1 985.)
318 R . D . Blandford
Devinney, 197 1) and BM Ori (Wilson, 1972) can all be modelled using
normal companion stars and so there is no obligation to postulate the
presence of a black hole. However, the absence of X-rays, only forbids an
accreting compact object and even here, the example of A0620-00 warns us
that it is possible for X-rays from a disk to be beamed away from us or
attenuated. It is still worth investigating other possibilities. Black hole
binaries that are currently X-ray active, may constitute only a small fraction
of the total.
Rapid ( ,.., 1 ms), aperiodic X-ray variability (e.g. Sunyaev, 1973) was
proposed as a signature of a black hole and for a long while Cir X- 1 was held
to be a black hole on these grounds. However, Cir X- 1 undergoes X-ray
bursting on timescales ,.., 10 s and this is strongly believed to be a signature
of a neutron star (Tennant, Fabian and Shafer, 1986). Conversely, the
known neutron star binary, V0332 + 52 (as well as GX339-4 and Cir X- 1)
exhibits rapid flickering, just like Cyg X- 1 (Stella et al. , 1985). It is rather
surprising that rapid variability has been regarded for so long as a peculiar
signature of a black hole as orbits near the surface of a ,.., 1 .4 M 0 , 10 km
neutron star have periods ,.., 1 ms, similar to those for orbits around a more
massive black hole.
Spectral criteria have also been proposed as an indication of the presence
of a black hole. In particular, White and Marshall ( 1984) pointed out that
the black hole candidates had unusually soft (i.e. steep) X-ray spectra.
However, these sources are not always in this state and many of them
(notably Cyg X- 1) change frequently to emit a hard (i.e. flat) X-ray spectrum.
Unfortunately Cir X- 1 behaves in a similar manner and so the usefulness of
this criterion is suspect at the moment.
As an illustration of the importance of obtaining a good X-ray position,
consider the case of the X-ray source OAO 1653-40 identified by Polidan et
al. ( 1978) with the supergiant binary V86 1 Seo (Wolff and Beichmann,
1979), on the basis of two X-ray eclipses separated by four times the known
binary period. If the identification had been correct, then the mass function
of0.5 M 0 and spectral type of the star would have implied a minimum mass
for the secondary star of 12.5 M 0 . However, Parmar et al. (1980) found that
the X-ray emission came from a nearby pulsating X-ray source and that
V86 1 Seo was undetectable.
A more direct indication of a black hole would appear to be the discovery
of an X-ray source with an X-ray luminosity exceeding the Eddington limit
for a star with mass equal to the Oppenheimer-Volkoff limit. SMC-X- 1
(which is at the known distance of the Magellanic clouds) might have been
just such a star with a 1-35 keV luminosity of 6 x 10 3 8 erg s - 1 (the

Eddington limit for a 4 M 0 star) and yet it has regular pulsations with
,...,
period of 0.7 s (Lucke et al. , 1976) indicative of a neutron star, so this

argument is also unreliable.
More satellite X-ray observations with (""' arcsecond) positional accuracy
will be necessary to find additional black hole binary candidates. It will be
particularly advantageous if these are used in conjunction with an all-sky X
ray survey instrument that can be used to locate X-ray transients. Further
discussion of X-ray binaries can be found in the review of McClintock ( 1986)
and the conference proceedings edited by Sanford, Laskarides and Salton
( 1982).
8 .3 .3 Other possibilities
Active galactic nuclei and binary X-ray sources are not the only possible
environments for astrophysical black holes; neither need black holes be
restricted to the stellar or the million to billion M 0 mass range. However,
there is as yet no observational evidence at all for any black holes outside
AGN and binary stars. This has not deterred theorists from speculating
further about the possibilities.
As reviewed by Blandford and Thome (1979), Shapiro and Teukolsky
( 1983), there are expected to be at least as many black holes formed singly as
are retained in binaries. (If the supernova explosion is spherically symmetric
and more than half of the system mass is lost, then a newly formed black hole
will escape from a binary.) If, for illustration, we assume that main sequence
stars with initial masses in excess of 10 M 0 evolve to give black holes, whilst
more numerous stars of intermediate mass evolve to give neutron stars, then
we can use the stellar mass function (e.g. Bahcall and Soneira, 1980) to relate
the number density of dead neutron stars to the number density of black
holes. Estimates of the birth rate of radio pulsars indicate that there are
between 108 and 4 x 108 defunct radio pulsars in the Galaxy (Lyne,
Manchester and Taylor, 1985), and we might expect a third as many black
holes to be formed. The nearest one would then be less than 5 pc away.
Unfortunately, our understanding of the details of supernova explosions
and mass loss in evolving stars, although it has progressed substantially, is
still not at the point where we can confidently estimate what fraction of
massive stars evolve through to give black holes (e.g. Bethe and Brown,
198 5). (This estimate is in any case complicated by the additional
uncertainties associated with mass transfer in binaries which may allow a
neutron star to accrete enough mass to drive it over the Oppenheimer-
320 R. D. Blandford
Volkoff limit.) Unfortunately, even if black holes are this common, the
nearby ones will only accrete very slowly in the interstellar medium, and will
be unobservably faint (e.g. Shapiro and Teukolsky, 1983). ·
From the dynamics of stars in the solar neighbourhood, we can estimate
the density of matter in the galactic disk. This is known as the Oort limit and
the stars that we can see account for slightly more than half of it. Black holes
formed as stellar remnants in the first generation of Galactic stars may
account for the difference ( "' 0 . 1 M 0 pc - 3 , Bah call, 1984). There is al so a
'missing mass' problem associated with the halo of our Galaxy and again,
there must be as much invisible mass as visible mass at the solar radius.
These may be primordial ( "' 105 M 0) black holes. However there are some
constraints. Lacey and Ostriker ( 1985) have proposed that the halo mass
comprise "' 106 M 0 black holes perhaps formed primordially from
isothermal fluctuations in a hot, baryon-density universe (e.g. Carr and
Rees, 1984). These holes would steadily heat the galactic disk and thereby
explain the observed relationship between disk thickness and stellar age.
Lacey and Ostriker demonstrate that their interactions with individual stars
can change the local stellar velocity distribution function to a form
somewhat similar to that measured. Dynamical friction would cause masses
this heavy to sink into the galactic centre over the lifetime of a galaxy. In fact
there is a problem in that far too much mass will be accreted in our Galaxy.
These authors therefore propose that when three holes accumulate in a
galactic nucleus a dynamical slingshot (e.g. Begelman, Blandford and Rees,
1980), will lead to the ejection of some or all of the black holes. Presumably,
when the hole settling rate is large enough, a more massive black hole can be
formed. A large cosmic density of "' 106 M0 black holes, which this proposal
would require, might produce some observable gravitational lens images. A
difficulty with this model is that interstellar gas accreting onto these black
holes might make them bright enough to have been detected. For instance,
making somewhat optimistic assumptions about the accretion rate,
McDowell ( 1985) has deduced, on the basis of their absence from I RAS
source catalogues and proper motion surveys, that halo holes must be less
massive than 1000 M 0 . Another possible problem has been discussed by
"'
Bahcall , Hut and Tremaine ( 1984) who have argued that the existence of
very wide binaries, that would otherwise be disrupted by dynamical
interaction with heavy stars, constrains the missing mass in the disk to
comprise objects less massive than 2 M 0 , presumably excluding black
holes. This may also constrain the masses of halo holes.
As reviewed in Blandford and Thome ( 1979), globular clusters are
roughly a hundred times more X-ray luminous per unit mass than the rest of
the galaxy and when this was discovered, several authors proposed that
"" 1000 M 0 black holes reside in their nuclei (e.g. Bahcall and Ostriker,
1975). However, when improved X-ray positions became available, it was
clear that the X-ray sources were not located at the cluster centres as would
be expected for a massive object. From the observed distribution of these
sources, it can be concluded that the ratio of the mass of the X-ray-emitting
object to the mass of the field stars (typically ""' 0.5 M0) lies in the range
"" 1 .5--5 (Grindlay, 1985). They are therefore believed to be accreting
neutron star binaries. Nevertheless, a problem may remain in that more
globular clusters should show evidence for collapse of their cores. Larson
( 1984) has proposed that in fact central cusps are present in many globular
clusters, but these comprise invisible black holes which sink to the bottom of
the potential well.
Discovery of six pairs of quasars that are possibly gravitationally lensed
(e.g. Turner, 1986), (three of which show no evidence for luminous lenses),
has revived interest in cosmological � 101 2 M 0 massive black holes (e.g.
Paczynski , 1986). These have the additional advantage that they create only
two images as generally observed to be the case, rather than an odd number
of images as one would expect with a non-singular transparent lens (e.g.
Narayan, 1986).
Planetary mass primordial black holes have been 'postulated by Freese,
Price and Schramm ( 1983). These would have masses between the Hawking
mass ( ......, 10 1 5 g) and 1 M 0 . They may stimulate galaxy formation.
8.4 Future prospects
Black holes, an inevitable consequence of general relativity, were predicted
long before there were any observational indications that they exist.
Although astronomers have treated them with some scepticism, their place
as endpoints of stellar and galactic nuclear evolution now seems fairly
secure. The arguments for the X-ray sources Cyg X- 1 , LMCX-3 and A0620-
00 (and possibly also LMCX- 1 and SS433), being black holes in mass
transfer binaries are at least as strong as (to take one example) the reasons
for believing that helium is made in the big bang, and, in both cases, rely
essentially on Newtonian dynamics, the assumption that locally discovered
physical laws operate elsewhere in the universe, and common sense. In
active galactic nuclei, there is now an impressive body of circumstantial
evidence that most of the power derives from a coherent spinning object of
size not a lot larger than its Schwarzschild radius. The efficiency with which
322 R. D. Blandford
this object converts mass into radiation must exceed - 10 - 3 and perhaps
also 10- 2 . A - 106-10 9 M 0 black hole fits this description c<:> mpletely. The
reason why we do not reject competitors such as spinars with the conviction
that we reject neutron stars in Cyg X- 1 is that spinars are less well-defined
physical entities, and we find it hard to describe their structure and examine
their stability properties. Perhaps Nature will have the same difficulties.
After all , we do not consider it reasonable that the X-ray-emitting object in
Cyg X- 1 be a rapidly spinning disk of gas, supported by turbulence and
magnetic field for, say - 10 1 5 rotational periods . So, why should we take
any more seriously the notion of a similar structure a million times larger in
AGN surviving for - 109 periods?
Of course none of this proves that black holes exist. As I have argued , a
rigorous proof is going to be extremely hard to come by. This is the normal
state of affairs in most sciences. It is surely more productive at this stage to
accept the evidence and proceed . To quote the third volume of the Principia
(Newton, 1687), 'In experimental philosophy we are to look upon
propositions inferred by general induction from phenomena as accurately or
very nearly true, notwithstanding any contrary hypothesis that may be
imagined , till such time as other phenomena occur, by which they may either
be made more accurate, or liable to exceptions.'
Let me give some examples of ways of proceeding that are actively being
pursued . Firstly, it is slightly surprising that so few stellar mass black holes
have been found ; far less than was expected in 1972 . By contrast, almost 500
neutron stars have been discovered since 1 967. Perhaps this is telling us
something of importance about mass loss in the final stages of stellar
evolution or the dynamics of supernova explosions and that the usual corpse
of a massive star is a neutron star, black holes requiring somewhat unusual
conditions for their formation. Secondly, black holes in binary systems
display some common, though not exclusive, properties such as rapid
variability, spectral changes etc. Perhaps this behaviour, although
improperly understood , can be isolated in AGN and used to infer the masses
of the black holes. Thirdly , SS433 shows features in common with some
extragalactic double radio sources. Have we looked hard enough for a true
extragalactic counterpart to SS433?
Excepting chance discoveries, the prospects for observational advance in
this field lie mostly with space astronomy. X-ray telescopes are crucial to
finding more stellar mass black holes and to probing the most rapid
variability in AGN . Further advances must await the launches of ROSAT
and AXAF. In addition, it will be a great disappointment if Space Telescope
does not revolutionise our view of AGN by probing the central stellar cusps,
finding optical jets and telling us what types of galaxy are associated with the
different forms of activity. In particular, understanding the 'environmental
impact' of an active nucleus on the surrounding galaxy should help us to
determine the duty cycle for this activity. Finally, Gamma Ray Observatory
may discover the high energy photons that should be emitted if electron
positron pair plasmas are a prominent feature of AGN.
On the theoretical front, the major obstacle to progress, has been our
ignorance of the details of the viscosity in accretion disks and of the
dynamical importance of large scale magnetic field. The best prospects for
progress here probably lie with large scale numerical computation. This will
be carried out in the form of experiments on the behaviour of simple systems
which should give further insight into the general properties of relativistic
magnetohydrodynamical flows. There is little prospect of performing
detailed simulations of sources to which observational data can be fitted.
Similar explorations of radiative transfer in high temperature plasmas are
underway and should be interpreted in the same spirit.
Although progress on the astrophysical theory of black holes has been
disappointingly slow, it is some comfort to recall that it took a hundred
years for the work of Euler, Lagrange and Laplace to reap the full benefit of
the scientific revolution brought about by the publication of the Principia
and over 150 years for its most spectacular observational success, the
discovery of Neptune. Perhaps we should be more patient in our attempts to
derive comparable results from the general theory of reliativity.
Acknowledgements
I thank the Chief, Division of Radiophysics, CSIRO, Sydney and the Acting
Director Anglo-Australian Observatory for hospitality during the writing of
this review. I also thank David Allen for helpful discussions on the Galactic
centre and Roger Romani for guidance on SS433. I am grateful to Ineke
Stacey for assistance with the preparation of the manuscript and Robyn
Shobbrook for bibliographic help. Support under the National Science
Foundation grant AST84- 1 5355 is gratefully acknowledged.
References
Abell, G. 0. and Margan, B. ( 1979). Nature, 279, 70 1 .

Abramowicz, M . A . , Jaroszynski, M . and Sikora, M . ( 1978). Astron. Astrophys., 63, 22 1 .
324 R. D. Blandford
Alcock, C. R. ( 1986). The Origin and Evolution of Neutron Stars, ed . D. Helfand. Reidel :
Dordrecht. (in press).
Arnett, W. D. and Bowers, R. L. ( 1977). Astrophys. J. Suppl., 33, 4 1 5 .
Arons, J . , Kulsrud , R . M . and Ostriker, J. P. ( 1 975). Astrophys. J . , 1 98 , 687.
Avni , Y. and Bahcall, J . N. ( 1975). Astrophys. J., 1 97 , 675 .
Baade, W. and Zwicky, F. ( 1934). Phys. Rev., 45, 1 3 8 .
Baade, W. and Minkowski, R . ( 1954). Astrophys. J., 206, 14.
Backer, D . and S ramek, R . ( 1987). In preparation.
Bahcall , J. N . , Hut, P. and Tremaine, S . D . ( 1984). Astrophys. J., 290, 1 5 .
Bahcall , J. N . and Ostriker, J. P . ( 1975). Nature, 256, 23 .
Bahcall , J . N . ( 1978). Ann . Rev. Astron . Astrophys., 16, 24 1 .
Bahcall , J . N . and . Soneira , R . M . ( 1980). Astrophys. J . Suppl., 44 , 73.
Bahcall , J . N . ( 1 984). Astrophys. J., 287, 926 .
Balick, B . and Brown , R . L. ( 1 974). Astrophys. J., 1 94 , 265.
Band, D. L. and Grindlay, J. E. ( 1985). Astrophys. J., 285, 702.
Bardeen, J . M . ( 1970). Nature, 226, 64.
Bardeen, J. ( 1973). Biack Holes, ed. B. DeWitt and B. S. DeWitt. Gordon and B reach :
New York.
Bardeen, J. and Wagoner, R. V. ( 1 969). Astrophys. J. Lett ., 158 , L65.
Bardeen, J. M . and Petterson, J. A . ( 1975). Astrophys. J. Lett., 1 95 , 65.
Barker, B . M. and O'Connell , R. F. ( 1 975) . Phys. Rev., D l2 , 329.
Barnes, J., Goodman, J. and Hut, P . ( 1 986). Astrophys. J., 300, 1 12.
Barr, R. and Mushotzky, R . F. ( 1986). Nature, 322, 42 1 .
Begelman, M . C , ( 1985). Astrophysics of Active Galaxies and Quasi Stellar Objects, ed .
J. Miller. University Science Books : California.
Begelman, M . C., Blandford, R. D. and Rees, M . J. ( 1 980). Nature, 287, 307.
Begelman, M . C. and Rees, M . J. ( 1978). Mon . Not. R. Astron . Soc., 188, 847.
Begelman, M . C., Blandford , R. D. and Rees, M. J. ( 1984). Rev. Mod. Phys., 56, 255.
Bethe, H. A. and Brown , G . ( 198 5). Sci. Amer., 252, 5, 40.
Bicknell, G. V . and Gingold, R. A . ( 1 986) . Preprint.
Bisnovatyi-Kogan, G. S. and Sunyaev, R . A . ( 1 972). Sov. Astron. A .J., 16, 206.
Blandford , R. D. and Thorne, K. S. ( 1 979). General Relativity, ed . S . W. Hawking and
W. Israel . Cambridge University Press : Cambridge.
Blandford, R. D. and Rees, M . J. ( 1974). Mon. Not. R. Astron. Soc., 169, 395.
Blandford, R . D. and Znajek, R. L. ( 1977). Mon. Not. R. Astron . Soc. , 179, 433 .
Bolton, C. T . ( 1972). Nature, 235, 27 1 .
Boyle, B . J . , Fong, R . , Shanks, T. and Peterson , B . A . ( 1987). Mon . Not. R . Astron. Soc.,
In press.
Braginsky, V . B., Caves, C. M . and Thome, K. S. ( 1977). Phys. Rev., D IS, 2047.
B regman, J., Butter, D., Kemper, B., Koski, A., K raft R . P . and Stone, R. P. S. ( 1 973).
Astrophys. J. Lett., 186, L 1 1 7 .
Brown , R. L. and Liszt, H . S . ( 1984). Ann. Rev. Astron. Astrophys., 2 2 , 223.
Burbidge, G. R . ( 1956). Astrophys. J., 129, 849 .
Burbidge, G. R . ( 1959). Paris Symposium on Radio Astronomy, ed . R . N . Bracewell.
Stanford : California.
Burbidge, G. R. ( 1964). Proceedings Solvay Conference.
Burbidge, E. M . , Burbidge, G. R. and Prendergast , K . H. ( 1959). Astrophys. J., 130 , 26.
Burbidge, E. M . , Burbidge, G. R . and Sandage, A . ( 1 964). Rev. Mod. Phys., 35, 97.
Cameron , A . G. W. ( 197 1). Nature, 229 , 178 .
Carr, B . J. and Rees, M . J. ( 1 984). Mon. Not. R. Astron. Soc., 206 , 3 15 .
Astrophysical black holes 32:
Carter, B . ( 1979). General Relativity. An Einstein Centenary Survey, ed . S. W. Hawking

and W. Israel. Cambridge University Press : Cambridge.
Carter, B. and Luminet, J . P . ( 1982). Nature, 296 , 2 1 1 .
Chandrasekhar, S . ( 193 1). Mon. Not. R . Astron . Soc., 91 , 456.
Chiu , H-Y . ( 1964). Ann . Phys., 26, 364.
Christodoulou, D. (1970) . Phys. Rev. Lett., 25, 1 596.
Clark , D . H . and Murdin, P . ( 1978). Nature, 276, 45.
Cohen , M. H . , Cannon, W . , Purcell, G . H., Shaffer, D . B., Broderick, J . J., Kellermann,
K . I . and Jauncey , D. L. ( 197 1). Astrophys. J., 170, 207.
Cohen , M . H . ( 1 986). Highlights of Modern Astrophysics: Concepts and Controversies,
ed . S. Shapiro and S . Teukolsky . Wiley : New York .
Colpi, M . , Maraschi , L . and Treves, P . ( 1 984). Astrophys. J., 280, 3 19.
Colgate, S . A . ( 1967). Astrophys. J., 150, 163.
Collins, G . W . I I ( 1985). Mon. Not. R. Astron. Soc., 213, 279.
Coroniti, F . V. ( 1985). Unstable Current Systems in Astrophysics, ed. Kundu and
Holman . Reidel : Dordrecht.
Cowley, A. P., Crampton, D., Hutchings, J . B . , Remillard , R. and Penfold, J. E. ( 1983).
Astrophys. J., 272, 1 18 .
Crampton , D . and Hutchings, J. B . ( 198 1). Astrophys. J . , 251 , 604.
Crawford, M . K . , Genzel , R . , Harris, A. I . , Jaffe, D. T . , Lacy, J. H . , Layten, J. B . ,
Serabyn, E . and Townes, C . H . ( 1985). Nature, 315, 467.
Damour, T. ( 1975). Ann. N.Y. Acad Sci., 263, 1 13 .
Damour, T. ( 1978). Phys. Rev., D I S , 3598.
David, L . P., Durisen , R . H. and Cohn, H. ( 1986). Astrophys. J., In press.
Davies, R. L. and Illingworth, G. ( 1986). Astrophys. J., 302 , 234.
de Vaucouleurs, G. and Nieto, J. L. ( 1979). Astrophys. J., 230, 697.
Devinney, E. J. ( 197 1). Nature, 233, 1 10.
D ressler, A . ( 1980). Astrophys. J. Lett. , 240, 4 1 .
D ressler, A. ( 1984). Astrophys. J . Lett., 286, L97 .
Duncan , M . and Shapiro, S . L . ( 1983). Astrophys. J., 268, 565.
Duncan, M. J. and Wheeler, J. C . ( 1 980). Astrophys. J. Lett., 237, L27 .
Dyson , J . (ed .) ( 1985). Active Galactic Nuclei, Manchester University Press : Manchester.
Eddington , A. S. ( 1935). The Observatory, 58, 3 8 .
Einstein , A . ( 1939). Ann . Math . Princeton, 40 , 922.
Fabian , A . C. and Rees, M . J. ( 1979). Mon. Not. R. Astron. Soc., 187, 13 p.
Fabian, A . C., Eggleton , P . P ., Hut, P . and Pringle, J . E. ( 1986). Astrophys. J., 305,
333.
Fath, E . A . ( 1908). Lick, Obs. Bull., 5, 7 1 .
Feigelson, E . D . , Bradt, M . , McClintock , J . , Remillard , R . , Urry, C . M . , Tapia, S . ,
Geldzahler, B., Johnston, K . , Romanishin, W . , Wehinger, P . A . , Wycoff, S . , Madejski,
G . , Schwartz, D. A . , Thorstensen, J. and Schaefer, B. E. ( 1986). Astrophys. J., 302,
337 .
Filippenko, A . L . and Sargent, W . L . W . ( 1985). As.trophys. J. Suppl., 57, 3.
Frank, J . , King, A . R . and Raine, D . J. ( 198 5). Accretion Power in Astrophysics.
Freese, K . , Price, R. and Schramm, D. ( 1983). Astrophys. J., 275, 405.
Giacconi, R . , Gursky, H . , Paolini, F. R. and Rossi, B. B . ( 1962). Phys. Rev. Lett., 9, 439 .
Gies, D . R. and Bolton, C. T. ( 1986). Astrophys. J., 304, 37 1 .
Ginzburg, V . N . and Ozernoi, L . M . ( 1964). Sov. Phys. JETP, 47, 1030.
Gold, T., Axford, W. I. and Ray, E. C. ( 1964). Quasi-stellar Sources and Gravitational
326 R. D. Blandford
Collapse, ed. Robinson , I., Schild, A., Schucking, E. L., Chicago University Press :
Chicago.
Giuricin, G., Mardirossian, F., Mezzetti, M. and Ramella, M . (eds.) ( 1986). Structure
and Evolution of Active Galactic Nuclei. Reidel : Dordrecht.
Goldreich , P., Goodman, J. and Narayan, R. ( 1986). Mon. Not. R. Astron. Soc. In press.
Greenstein , J. and Schmidt, M . ( 1 964). Astrophys. J., 140, 16 1 .
Grindlay, J. ( 1985). Dynamics of Star Clusters, ed . J. Goodman and P . Hut. Reidel :
Dordrecht.
Gunn, J. E. and Ostriker, J . P . ( 1970). Astrophys. J., 157, 1 395.
Gurzadyan, V . G. and Ozemoi , L. M . ( 198 1). Astron. Astrophys., 95, 39.
Hall, D . N. B., Kleinmann, S. G. and Scoville, N. Z. ( 1982). Astrophys. J. Lett., 260,
L53.
Harrison, B. K., Thorne, K. S., Wakano , M. and Wheeler, J. A . ( 1965). Gravitational
Theory and Gravitational Collapse. Chicago University Press : Chicago .
Hawking , S . W. ( 1 965). Phys. Rev. Lett., 15, 689.
Hawley, J. ( 1986). Preprint .
Hazard, C . , MacKey, M . B. and Shimmins, A. J. ( 1963). Nature, 1 97, 1037.
Hewish, A., Bell , S. J., Pilkington, J. D. H., Scott , P. F. and Collins, R. A. ( 1 968).
Nature, 2 17, 709.
Hoyle, F. and Fowler, W. A. ( 1963a). Mon . Not. R. Astron. Soc., 1 25, 169.
Hoyle, F. and Fowler, W. A. ( 1963b). Nature, 197, 533.
Hutchings, J. B., Crampton, D. and Cowley, A. P. ( 1983). Astrophys. J. Lett. , 275, L43 .
Iben, I. ( 1 964). Quasi-stellar Sources and Gravitational Collapse, ed. Robinson , I . ,
Schild, A. and Schucking, E . L. Chicago University Press : Chicago.
Jennison, R. C. and Das Gupta , M. K. ( 1953). Nature, 1 72, 96 .
Katz, J. I . , Anderson, S . F . , Margon, B. and Grandis, S. A . ( 1982). Astrophys. J., 260,
780.
Keel , W. ( 1985). Astrophysics of Active Galaxies and Quasi-stellar Objects, ed. J . M iller.
University Science Book s : Mill Valley, California.
Kerr, R. P. ( 1963). Phys. Rev. Lett. , 1 1 , 237.
Koo , D. C. ( 1986). Structure and Evolution of Active Galactic Nuclei, eds . G. Giuricin,
F. Mardirossion, M. Mezzetti and M . Ramella. Reidel : Dordrecht.
Kormendy, J. ( 1 986). (Preprint).
Krolik, J. H. and London, R . A. ( 1 983). Astrophys. J., 267, 37 1 .
Kumar, S . and Pringle, J. E. ( 1985). Mon. Not. R . Astron. Soc., 213, 435.
Lacey, C. G. and Ostriker, J. P. ( 1 975). Nature, 256, 23.
Landau , L. ( 1932). Phys. Z. Sowjetunion. , l, 285.
Laplace, P . S . ( 1795). Le System du Monde, Vol . II. Paris.
Larson , R. B. ( 1984). Mon. Not. R. Astron. Soc., 2 10 , 763 .
Le Blanc, J. M . and Wilson, J. R . ( 1969). Astrophys. J., 161 , 54 1 .
Leibowicz, E . ( 1984). Mon. Not. R . Astron . Soc . , 210, 279.
Lense, J. and Thirring, H. ( 1 9 1 8). Phys. Z . , 19, 156.
Lightman, A. P. and Shapiro, S . L. (1977) . Astrophys. J., 2 1 1 , 244.
Lin, D. N. C., Papaloizou , J. C . B. and Faulkner, J. ( 1985). Mon . Not. R . Astron . Soc.,
2 1 2 , 105.
Lingenfelter, R. E. and Ramaty, R. ( 1982). The Galactic Centre, ed . G. Riegler and
R . Blandford . American Institute of Physics, New York .
Lo, K-Y., Backer, D . C . , Ekers, R . D . , Kellermann , K . I . , Reid , M . and Moran , J. M .
( 198 5) . Nature, 315, 124.
Lo, K-Y . ( 1 986). Science, 233, 1 394 .
Longair, M . S., Ryle, M . and Scheuer, P . A. G . ( 1973). Mon. Not. R. Astron. Soc. , 164,
243 .
Lucke, R . , Yentis, D . , Friedman, H . , Fritz, G . and Shulman , S . ( 1976). Astrophys. J.
Lett., 206, L25 .
Lynden-Bell , D . ( 1969). Nature, 223, 690.
Lynden-Bell , D . ( 19 7 1 ) . Nuclei of Galaxies, ed. D . J . K . O'Connell. North Holland :
Amsterdam .
Lynden-Bell , D . and Rees, M . J . ( 197 1). Mon . Not. R. Astron. Soc. , 152 , 46 1 .
Lynden-Bell , D . ( 1978). Phys. Scripta, 17, 185.
Lyne, A . G . , Manchester, R . N . and Taylor, J . H . ( 1985). Mon. Not. R. Astron. Soc.,
213, 6 13 .
M acDonald, D . A . and Thorne, K . S . ( 1982). Mon. Not. R. Astron. Soc. , 198, 345 .
McClintock, J . E . and Remillard, R . A . ( 1 986). Astrophys. J., 308 .
McClintock, J . E. ( 1986) . Preprint.
McDowell, J. ( 1985). Mon. Not. R. Astron. Soc. , 217, 77.
Maran , S . P. and Cameron, A . G . W . ( 1964) . Physics of Non-thermal Radio Sources.
NASA : Washington .
Margon , B . , Ford, H . C., Katz, J . I . , Kwitter, K . B . , Ulrich , R . K . , Stone, R . P . S. and
Klemola, A . ( 1 979). Astrophys. J. Lett., 230, L4 1 .
Margon , B . ( 1980). Sci. Amer., 243, 4 , 44.
Margon , B. ( 1984). Ann . Rev. Astron. Astrophys. , 22, 507.
Margon , B . , Bowyer, S. and Stone, R. S. ( 1973) . Astrophys. J. Lett., 185, L l l3 .
M arshall, F. E., Holt, S . , Mushotzky, R . F. and Becker, R . M . ( 1983). Astrophys. J .
Lett., 269, L3 l 1 .
Matthews, T . A . , M altby, P . and Moffet, A . T . ( 1962). Astrophys. J., 137 , 1 5 3 .
M azeh, T . , von Paradis, J . , van der Heuvel, E . P . J . and Savonije, G. J . ( 1986). Astron.
Astrophys., 157, 1 1 3 .
Michell, J . (1 784). Philos. Trans., 74, 3 5 .
Milgrom, M . ( 1 979). Astron. Astrophys., 76, L3 .
Miller, J . (ed.) ( 1 985). Astrophysics of Active Galaxies and Quasi-Stellar Objects,
University Science Books : California.
Misner, C. W., Thorne, K. S. and Wheeler, J. A. (1973). Gravitation. Freeman :
San Francisco.
Moffet , A . T . , Gubbay, J . , Robertson, D . S . and Legg, A . J . (197 1 ). External Galaxies
and Quasi-stellar Objects, ed. D. S. Evans. Reidel: Dordrecht.
Narayan , R. ( 1986). Quasars, ed. G. Swarup and V. Kapahi. Reidel : Dordrecht.
Newsom, G . H. and Collins , G . W . II ( 1 982). Astrophys. J., 262, 7 14.
Newton , A . J . and Binney, J . ( 1984) . Mon. Not. R. Astron . Soc., 210, 7 1 1 .
Norman, C . A . and Silk, J . ( 1983). Astrophys. J., 266, 502.
Novikov, I . D. and Thorne, K. S. ( 1973). Black Holes, ed. B. DeWitt and B. S. DeWitt.
Gordon and Breach : New York.
Oppenheimer, J. R. and Volkoff, G. M . ( 1939) . Phys . Rev., 55, 374.
Oppenheimer, J. R. and Snyder, H. ( 1939) . Phys. Rev., 56, 455.
Ozernoi, L. M. and Chertoprid, V . E. ( 1969). Sov. Astr. A. J., 46, 940.
Pacholczyk, A . G . and Weymann, R. E. ( 1968). Astron. J., 73, 8 50.
Paczynski, B . ( 1974). Astron. Astrophys. , 34, 16 1 .
Paczynski, B . ( 1984). Astrophys. J. Lett. , 273, L8 1 .
Paczynski, B . ( 1986). Nature, 321 , 4 19.
Paczynski, B. and Wiita, P. J . ( 1 980). Astron. Astrophys., 88, 23.
Papaloizou, J . C. B . and Pringle, J . E. ( 1984) . Mon. Not. R. Astron. Soc., 208, 72 1 .
328 R. D. Blandford
Papaloizou, J . C. B . and Pringle, J. E. ( 1985). Mon. Not. R. Astron. Soc. , 213, 799 .
Pakull, M . W. and Angebault, L. P . ( 1986) . Nature, 322 , 5 1 1 .
Parmar, A . N . , Brandiardi-Raymont, G., Pollard , G . S . G . , Sanford, P . W., Fabian ,
A . C., Stewart, G . C., Schreier, E. J . , Poliden, R . S . , Oegerle, W . E. and Locke, M .
( 1980). Mon . Not. R. Astron. Soc., 193, 49p.
Penrose, R . ( 1965). Phys. Rev. Lett., 14, 57.
Penrose, R. ( 1969). Rev. Nuovo. Cimento, 1, 252.
Phinney, E. S. ( 1983). Unpublished thesis, University of Cambridge .
Polidan, R . S., Pollar, G. S . G . , Sanford, P . W . and Locke, M . C. ( 1978). Nature, 275,
296.
Press, W. H. ( 1972). Astrophys. J., 138, 2 1 1 .
Pringle, J. E . ( 198 1). Ann. Rev. Astron . Astrophys., 19, 137.
Pudritz, R . E. ( 1983). Mon. Not. R. Astron . Soc., 195, 88 1 .
Reber, G. ( 1944). Astrophys. J . , 100, 279.
Rees, M . J. ( 1966). Nature, 2 1 1 , 468 .
Rees , M . J . ( 197 1). Nature, 229, 3 12.
Rees, M. J . ( 1977). Ann. NY Acad. Sci. , 302, 6 1 3 .
Rees , M . J . ( 1978). Nature, 275, 5 16.
Rees, M. J . ( 1982). The Galactic Centre, ed . G . Riegler and R . D . Blandford . American
Institute of Physics : New York.
Rees, M . J. ( 1984). Ann. Rev. Astron. Astrophys., 22, 47 1 .
Rhoades, C. E . and Ruffini , R . ( 1 974) . Phys. Rev. Lett., 33, 324.
Richstone, D. and Tremaine, S . D . ( 1985). Astrophys. J., 296, 370.
Robinson, I . , Schild, A . and Schucking, E. ( 1964). Quasi-stellar Sources and Gravitational
Collapse. Chicago University Press : Chicago.
Roos, N. ( 198 1). Astron. Astrophys. , 104, 2 1 8 .
Sakimoto, P . J . and Coroniti, F. V . ( 198 1). Astrophys. J., 247, 19.
Salpeter, E. E. ( 1964). Astrophys. J., 140, 796 .
Salpeter, E. E. ( 197 1). Nature Phys. Sci., 233, 5 .
Salpeter, E. E . and Wagoner, R . V . ( 197 1). Astrophys. J., 164, 557.
Sanders, R . H. ( 1970). Astrophys. J., 162 , 784.
Sanders, R . H . and Allen , D . A . ( 1986) . Nature, 319, 19 1 .
Sanford, P . W . , Laskarides, P . and Salton, J . eds. ( 1982). Galactic X-ray Sources. Wiley :
London .
Sargent, W. L . W., Young, P . J . , Boksenberg, A . , Shortridge, K . , Lynds, C. R . and
Hartwick, F. D. A. ( 1978). Astrophys. J., 221 , 7 3 1 .
Scheuer, P . A. G . ( 1974) . Mon. Not. R . Astron. Soc., 166, 5 1 3 .
Schmidt , M . ( 1963). Nature, 197, 1040.
Schultz, A. L. and Price, R. H. ( 1985). Astrophys. J., 291 , 1 .
Schreier, E . , Gursky, H . , Kellogg, E . , Tananbaum, H . , Giacconi, R . ( 197 1). Astrophys. J .
Lett., 1 70, L2 1 .
Seyfert, C. K . ( 1943). Astrophys. J . , 97, 28.
Shakura, N . and Sunyaev, R . A . ( 1973). Astron . Astrophys., 24, 337.
Shapiro, S . L. and Teukolsky , S . A. ( 1983). Physics of Compact Objects: Black Holes,
White Dwarfs and Neutron Stars. Wiley : New York.
Shapiro, S . L. and Teukolsky. S . A. ( 1985). Astrophys. J. Lett., 292, L4 1 .
Shields, G. A . , McKee, C . F . , Lin , D . N . C. and Begelman , M . C. ( 1986) . Astrophys. J.,
306, 90.
Shklovsky, I . S. ( 1967). Astrophys. J. Lett., 148, L l .
As t rophysical black holes 329
Slipher, V. ( 19 17). Lowell Observatory Bulletin, 3, 59.

Smith, H . J . and Hoffieit, D. ( 1963). Nature, 198, 650 .
Soltan, A . ( 1982). Mon . Not. R . Astron. Soc . , 200, 1 15.
Spitzer, L . and Saslaw , W . C. ( 1966). Astrophys. J., 143, 400.
Stella , L . , White, N . E., Davelaar, J . , Parmar, A . N . , Blissett, R . J . and va� de Klis, M .
( 1985). Astrophys. J . Lett., 288, L45.
Stoeger, W . R . and Pacholczyk, A . G . ( 1983). Proc. 3rd Marcel Grossmann Meeting, ed.
Hu, N .
Sturrock , P . A . ( 1965). Nature, 205, 86 1 .
Sunyaev, R . A . ( 1973). Sov. Astron., 16, 94 1 .
Svennson, R . ( 1986). Radiation Hydrodynamics and Compact Objects, ed . D . M i halas and
K-H. A . Winkler. Springer : Berlin.
Swamp, G . and Kapahi , V . (eds.). ( 1 986). Quasars. Reidel : Dordrecht.
Tennant, A . F., M ushotzky, R . F., Boldt, E. A . and Swenk, J . H. ( 198 1). Astrophys. J.,
251 , 1 5 .
Tennant, A . F., Fabian, A . C. and S hafer, R . A . M . ( 1986). Mon . Not. R. Astron. Soc.,
219, 87 1 .
Thorne, K . S . ( 1966). In Proc. Intern . School Phys. 'Enrico Fermi' Course 35, Varenna,
Italy, ed . L. Gratton .
Thorne, K . S . , Price, R . H . and MacDonald , A. ( 1976) . Black Holes: The Membrane
Paradigm. Yale University Press : New Haven .
Trimble, V . L . and Thorne, K . S . ( 1969). Astrophys. J., 1 56 , 10 1 3 .
Trimble, V . L ., Rose, W . K . and Weber, J . ( 1973). Mon. Not. R . Astron. Soc . , 162, lp.
Tonry, J . ( 1984) . Astrophys. J. Lett., 283, L27.
Turner, E. L . ( 1986) . Dark Matter in the Universe, ed . G. R . Burbidge. Reidel :
Dordrecht.
Wald , R. M . ( 1974). Phys. Rev., DlO, 1680.
Webster, B. L. and Murdin, P . ( 1972). Nature, 235, 37.
Wheeler, J . A. ( 1966). Ann. Rev. Astron . Astrophys., 4, 393.
Wheeler, J . A . ( 1968). American Scientist, 56, 1 .
W hite, N . E . and Marshall , F . E . ( 1984). Astrophys. J . , 281 , 349 .
Whitney , A . R . , Shapiro, I . I . , Rogers, A . E. E. , Robertson, D . S . , Knight, D . A., Clark,
T. A . , Goldstein , R . M ., Marandino, G. E. and Vandenberg, N . R. ( 1 97 1). Science,
173, 225.
Wilson, R. E. ( 1972). Astrophys. Sp. Sci., 19, 165.
Wolff, S . C. and Beichmann, C. A . ( 1979) . Astrophys. J., 230, 5 19.
Woltjer, L . ( 1959) . Astrophys. J., 1 30 , 3 8 .
Woltjer, L . ( 197 1). Nuclei of Galaxies, ed . D . J . K . O'Connell. North Holland :
Amsterdam.
Young, P . H . , Westphal, J . A . , Kristian , J . , Wilson, C. P . and Landauer, F. P . ( 1978).
Astrophys. J., 221 , 72 1 .
Zeldovich, Y a B., ( 1964a). Sov. Phys. Dok/., 9 , 195.
Zeldovich, Ya B. and Guseynov, 0 . H. ( 1965). Astrophys. J., 144, 84 1 .
Zeldovich, Y a B. and Novikov, I . D . ( 1964b). Sov. Phys. Dok!., 158, 8 1 1 .
Zeldovich, Ya B. and Novikov, I. D . ( 1965) . Sov. Phys. Usp., 7, 763 .
Zeldovich, Ya B. and Novikov, I . D. ( 1966). Sov. Phys. Usp., 8, 522 .
Znajek, R . L. ( 1978). Mon. Not. R . Astron. Soc., 185, 833.
9
Gravitational radiation
K I P S . T H O RN E t
9. 1 Introduction
9 .1 . 1 The motivations for gravitational-wave research

The discovery of cosmic radio waves in the 1930s and their detailed study in
the 40s, 50s, and 60s created a revolution in our view of the universe
(Kellerman and Sheets, 198 3 ; Sullivan, 1982, 1 984). Previously the universe,
as viewed by light, was regarded as serene and quiescent - dominated by
stars and planets that wheel smoothly in their orbits, shining steadily and
evolving (with few exceptions) on timescales of millions or billions of years.
By contrast, the universe as viewed by radio waves was violent : galaxies in
collision, jets ejected from galactic nuclei, quasars with luminosities far
greater than our galaxy varying on timescales of hours, pulsars with
gigantic radio beams rotating at several or many rotations per second ; these
were the typical strong radio emitters.
The radio revolution was so spectacular because the information carried
by radio waves is so different from that carried by light. A factor 107
difference in wavelength meant the difference between photons
predominantly thermal in origin (light) and photons predominantly
nonthermal (radio), the difference between the bremsstrahlung and atomic
transitions of stellar and planetary atmospheres on one hand, and the
synchrotron radiation of intergalactic magnetized plasmas on the other.
As different as cosmic radio and optical radiations may be, their
differences pale by comparison with those of electromagnetic waves and
gravitational waves : cosmic gravitational waves should be emitted by, and
carry detailed information about, coherent bulk motions of matter (e.g.,
collapsing stellar cores) or coherent vibrations of spacetime curvature (e.g.,
t Supported i n part b y the National Science Foundation (AST-8 5 149 1 1).
Gravitational radiation 33 1
black holes). By contrast, cosmic electromagnetic waves are usually

incoherent superpositions of emission from individual atoms, molecules,
and charged particles. Gravitational waves are emitted most strongly in
regions of spacetime where gravity is relativistic and where the velocities of
bulk motion are near the speed of light. But electromagnetic waves come
almost entirely from weak-gravity , low-velocity regions, since strong
gravity regions tend to be obscured by surrounding matter. Gravitational
waves pass through surrounding matter with impunity, by contrast with
electromagnetic waves which are easily absorbed and scattered, and even by
contrast with neutrinos which, although they easily penetrate normal
matter, should scatter thousands of times while leaving the core of a
supernova.
These differences make it likely that, if cosmic gravitational waves can be
detected and studied, they will create a revolution in our view of the universe
comparable to or greater than that which resulted from the discovery of
radio waves.
It might be argued that we are now so sophisticated and complete in our
electromagnetically based (radio, millimeter, infrared, optical, ultraviolet,
X-ray, gamma-ray, cosmic ray) understanding of the universe, compared to
the optically based astronomers of the 1930s and 1940s, that a gravitational
wave revolution will be far less spectacular than was the radio revolution. It
seems unlikely to me that we are so sophisticated. I am painfully aware of
our lack of sophistication when I contemplate the sorry state of present
estimates of the gravity waves bathing the earth (Section 9 .4 below) : for each
type of gravity-wave source that has been studied , with the exception of
binary stars and their coalescences, either (i) the strength of the source's
waves for a given distance from earth is uncertain by several orders of
magnitude; or (ii) the rate of occurrence of that type of source, and thus the
distance to the nearest one, is uncertain by several orders of magnitude; or
(iii) the very existence of the source is uncertain.
Although these uncertainties make us unhappy when we try to plan for
the design and construction of gravitational-wave detectors, we will be
rewarded with great surprises when gravity waves are ultimately detected
and studied : the waves will give us extensive information about the universe
that we are unlikely ever to obtain in' any other way.
Detailed studies of cosmic gravitational waves are also likely to yield
experimental tests of fundamental laws of physics which cannot be tested in
any other way .
The first discovery of gravitational waves would directly verify the
332 K. S. Thorne
predictions of general relativity, and other relativistic theories of gravity,

that such waves should exist. (There has already been an indirect
verification, in the form of the observed inspiral of the binary pulsar due to
gravitational-radiation reaction ; Weisberg and Taylor, 1984; Taylor,
1987.)
By comparing the arrival times of the first bursts oflight and gravitational
waves from a distant supernova, one could verify general relativity's
prediction that electromagnetic and gravitational waves propagate with the
same speed - i.e., that they couple to the static gravity (spacetime curvature)
of our Galaxy and other galaxies in the same way. For a supernova in the
Virgo cluster of galaxies ( 1 5 Mpc distant), first detected optically one day
after the light curve starts to rise, the electromagnetic and gravitational
speeds could be checked to be the same to within a fractional accuracy
(l light day)/( 15 Mpc) = 5 x 10 - 1 1 •
By measuring the polarization properties of the gravitational waves, one
could verify general relativity's prediction that the waves are transverse and
traceless - and thus are the classical consequences of spin-two gravitons
(Eardley, Lee and Lightman, 1973 ; Eardley et al. , 1973).
By comparing the detailed wave forms of observed gravitational wave
bursts with those predicted for the coalescence of black-hole binaries (which
will be computed by numerical relativity in the next few years, see Section
9.3.3(e) below), one could verify that certain bursts are indeed produced by
black-hole coalescences - and, as a consequence, verify unequivocally the
existence of black holes and general relativity's predictions of their behavior
in highly dynamical circumstances. Such verifications would constitute by
far the strongest test ever of Einstein's laws of gravity.
9.1 .2 A brief history of gravitational-wave research
Einstein ( 19 16) laid the foundations of gravitational-wave theory within
months after his final formulation of general relativity - restricting himself to
weak (linearized) waves emitted by bodies with negligible self-gravity and
propagating through flat, empty spacetime. During the next few years
Einstein (1918) , Weyl ( 1 922), and Eddington ( 1 924) elaborated on Einstein's
initial work so that by the mid- 1920s the linearized theory of gravitational
waves was fully understood . However, it was clear - at least to Eddington
(1924) - that for sources with significant self-gravity (e.g. binary systems),
the linearized analysis was invalid.
Landau and Lifshitz ( 1 94 1 ) gave the first fairly satisfactory treatment of
the emission of waves by self-gravitating systems ; but a series of failed
Gravitational radiation 33�
attempts to analyze radiation reaction in such systems in the late 1940s and
1950s (pp. 73 and 74 of Damour, 1983) shook physicists' faith in the ability
of the waves to carry off energy, and even in the correctness of the Landau
Lifshitz formula for the emitted wave field. It required a clever thought
experiment by Bondi ( 1957) to restore faith in the energy of the waves, and a
series of beautiful and rigorous studies of the asymptotic properties of the
waves at infinity by Bondi and collaborators (Bondi, 1960; Bondi, van der
Burg, and Metzner, 1962 ; Sachs, 1962, 1963 ; Penrose, 1963a,b) and of the
propagation of short-wavelength waves through a curved background
spacetime by Isaacson ( 1968a,b) to restore faith that the fundamental theory
of gravitational waves is soundly based.
The experimental search for cosmic gravitational waves was initiated by
Joseph Weber ( 1960) at a time when almost nothing was known about
possible cosmic sources and when nobody else had the vision to see that
there were technological possibilities of ultimate success. After a decade of
effort, Weber ( 1969) announced to the world tentative evidence that his
resonant-bar gravity-wave detectors - one near Washington, DC, the other
near Chicago - were being excited simultaneously by gravitational waves.
There followed a six-year period of excitement and feverish effort as 1 5 other
research groups around the world tried to construct and operate similar bar
detectors (Tyson and Giffard, 1978 ; Amaldi and Pizella, 1979 ; de Sabbata
and Weber, 1977 ; Weber, 1986, and references therein). Sadly, even with
markedly improved sensitivities, these efforts gave no convincing evidence
that gravity waves were actually being seen.
In parallel with this experimental effort, astrophysicists worldwide
struggled through the early 1970s to milk, from electromagnetic
observations of the universe and from fundamental theory, as much
information as possible about the characteristics of the gravity waves that
might be bathing the earth. By the mid- 1970s a fuzzy but helpful picture had
begun to emerge : While the sensitivities of the detectors to kilohertz
frequency bursts arriving, say, three times per year had improved during the
early 70s from dimensionless amplitude h31yc "" 1 x 10 - 1 5 to h31 yc "" 3 x 10 - 1 6
(a factor 10 improvement i n energy flux), it seemed highly unlikely that such
bursts bathing the earth would exceed h3;yr "" 1 x 10 - 1 6 ; a reasonable
probability of success would require h31 yr "" 10 - 2 0 or better; and a high
probability would require h3/yr "" 10 - 21 to h3/yr "" 10 - 22 (Smarr, ed., 1979;
Fig. 9.4 below). Although these estimates were discouraging, the theoretical
efforts that produced them were making clear the enormous potential payoff
that could follow the successful detection of gravity waves.
334 K. S. Thorne
Fortunately, the experimental efforts of the early 1970s had pointed the
way toward major possible detector improvements; and, consequently,
although most of the first-generation experimental groups became
discouraged and dropped out, a handful of highly talented groups continued
onward into the 1980s with a second-generation effort involving major
technological changes such as cooling the bars to liquid-helium
temperatures, changing bar materials, switching from passive to active
transducers, and even developing completely new types of detectors, most
notably laser-interferometer gravity-wave detectors (called 'beam' detectors
in this chapter). These second-generation efforts have reached fruition in the
last few years : bars with kilohertz burst sensitivities h J/y r ,..., 10 - 1 7 (30 times
higher in amplitude than the first-generation and 1000 higher in energy) are
now collecting data in coordinated searches (Section 9.5.2(d) below); and
small-scale beam detectors with h3;y r "' 5 x 10 - 1 7 are now operating (Section
9.5.3(d) below) as prototypes for full-scale detectors with projected ultimate
sensitivities in the 10 - 22 region (Section 9.5.3(g) below). The regime of
possible success has been reached, and the regimes of reasonably probable
success and highly probable success look reachable - though only with
vigorous continuing efforts and the expenditure of non-trivial sums of
money.
In parallel with these 1980s' second-generation efforts, theorists have
redoubled their struggle to firm up our understanding of the waves bathing
the earth, but with only modest results : the problem of knowing what kinds
of sources actually occur, and how frequently, is hampered by the paucity
of electromagnetic information; and, as a result, the apparent recent
improvements in our knowledge (Section 9.4 below) might be little more
than changes of fashion . On the other hand, given a specific scenario for how
a postulated source behaves, theorists have become far more adept than
before - thanks not least to supercomputers - at computing the details of the
gravitational waves it should emit (Section 9.3.3 below). As a consequence,
when waves are ultimately detected, the prospects have become reasonable
for deciphering from them the details of their sources.
While the present, 1987, gravity-wave searches might bring success, it is
not likely they will. Thus, we must anticipate a continued vigorous effort at
technology development during the coming years, with the prospects of
success improving significantly at each step along the way. In parallel, we
must anticipate a continuing major effort by relativity theorists to refine
their ability to decipher the source behaviors corresponding to postulated
gravitational-wave forms, and a continuing effort by astrophysicists to give
Gravitational radiation 335
better guidance as to what kinds of sources actually exist and in what

numbers. The efforts must be great ; but jointly they are likely to give an
extremely valuable payoff.
9 . 1 .3 Overview of this chapter
This chapter reviews all aspects of gravitational-wave research -
experimental , theoretical relativity, and theoretical astrophysics. This
review is intended to be readable by physicists (including advanced
students) who are not specialists in general relativity, in experimental
gravity, or in astrophysics.
A word of warning : in preparing this review I have not done a thorough
literature search (I lacked the necessary energy !) ; nor have I cited all major
original references of which I am aware (that would have made the reference
list even longer than it is!). However, I have attempted to present all
significant ideas and issues with which I am familiar, citing wherever
possible the earliest occurrence of the idea or issue and one or more recent,
thorough discussions of it.
This review is divided into four major parts : the physical and
mathematical description of gravitational waves (Section 9.2), the
generation and propagation of gravitational waves (Section 9.3),
astrophysical sources of gravitational waves (Section 9.4), and the detection
of gravitational waves (Section 9.5).
The physical and mathematical description of gravitational waves
(Section 9.2) is presented in a form that compactifies and updates the
corresponding material in the textbook that I coauthored fourteen years ago
(Misner, Thome, and Wheeler, 1973 ; cited henceforth as MTW). Emphasis
focuses on the 'shortwave approximation' as a tool for defining waves
mathematically (Section 9.2 . 1), and on measurements in the proper
reference frame of an observer as a tool for defining waves physically
(Section 9.2.2). A special 'TT coordinate system' is then introduced (Section
9.2.3) for use in analyzing systems large compared to a wavelength of the
waves; and the energy, momentum, and quantization of gravitational waves
are discussed (Section 9.2.4).
The theory of the generation and propagation of gravitational waves
(Section 9.3) is far more sophisticated today than when MTW was written :
we now understand how, in realistic situations, to split the problem of wave
generation off from that of wave propagation, and how to mesh generation
and propagation together using the technique of 'matched asymptotic
expansions' (Section 9.3 . 1). We also understand more deeply the most
336 K. S. Thorne
elementary of all ways to compute wave generation, the 'quadrupole

formalism', and its relationship to radiation reaction in the emitting system
(Section 9.3.2). Unfortunately, the most strongly emitting systems should
violate the mathematical approximations that underlie the quadrupole
formalism and thus can be analyzed only in rough order of magnitude using
it; for more accurate analyses one must use a more sophisticated wave
generation formalism. A catalog of more sophisticated formalisms is given in
Section 9.3.3 - which is an updated version of a review I wrote ten years ago
(Thome, 1977). The theory of the propagation of the waves from source to
earth is sketched in Section 9.3.4, and various wave-propagation effects
(absorption , scattering, dispersion, tails, gravitational focusing, diffraction,
parametric amplification by background curvature, non-linear coupling of
waves to themselves, and generation of background curvature by the waves'
energy and momentum) are described in Section 9.3.5 - which with 9.3.4 is a
shortened version of my recent (Thome, 1983), long review of wave
propagation. Section 9.3 concludes with very brief descriptions of elegant
mathematical work on idealized wave-propagation situations : the
asymptotic structure of waves propagating toward 'future null infinity' in an
asymptotically flat spacetime (Section 9.3.6), and exact, analytic solutions to
the Einstein equations for wave generation and propagation (Section 9.3.7).
It is nearly a decade since a detailed and thorough review has been written
of astrophysical sources and detectors for gravitational waves (Smarr, ed.,
1979; Douglass and Braginsky, 1979) ; and in the intervening time these
topics have changed enormously. (For recent reviews of a number of
subtopics see the chapters in Deruelle and Piran, 1983.) Sections 9.4 and 9.5
attempt a comprehensive review in a manner that closely ties the sources to
the detection efforts. In these sections the sources and the detection
strategies are split up into three categories : those for gravitational-wave
'bursts' (Section 9.4. 1 and Fig. 9.4); those for periodic gravitational waves
(Section 9.4.2 and Fig. 9.6) ; and those for a stochastic gravitational-wave
background (Section 9.4.3 and Fig. 9.7). All previous reviews have been
cavalier about factors of 2 in the definitions of wave strengths and detector
sensitivities. This review tries to standardize the definitions, including
factors of 2. The standardization is based on signal-to-noise-ratio analyses
that are given at the beginnings of Sections 9.4. 1 , 9.4.2, and 9.4.3. For each
source that looks favorable for wave detection, Section 9.4 gives a
description of our current state of knowledge of the source and gives current
estimates of the wave strengths and other wave characteristics.
The burst sources treated in Section 9.4. 1 include supernovae (collapse of
a normal stellar core to form a neutron star, subsection c), the collapse of a
star or star cluster to form a black hole (subsection d), the inspiral and
coalescence of compact binaries (neutron stars and black holes, subsection
e), and the fall of stars and small holes into supermassive holes (subsection f).
Because our knowledge of sources is so poor, it is useful to estimate how
strong the strongest wave bursts bathing the earth could be without
violating our cherished beliefs about the laws of physics and the nature of the
universe; this is done in subsection g. The periodic sources treated in Section
9.4.2 include rotating neutron stars (rigidly rotating pulsars, and neutron
stars spun up by accretion until they encounter a radiation-reaction-driven
instability, subsection b ), and binary stars (including unevolved binaries,
WUMa stars, white-dwarf binaries, and neutron-star binaries, subsection
c). The stochastic sources in Section 9.4.3 include large numbers of binary
stars whose waves superpose stochastically (subsection b ) pre-galactic,
,
Population III stars (subsection c), the big-bang singularity in which the
universe began - with subsequent parametric amplification of its waves by
background curvature in inflationary and other scenarios (subsection d),
phase transitions in the subsequent but still early universe (subsection e),
and cosmic strings produced by phase transitions (subsection f). Present
estimates of the strengths of the waves from all these sources are shown in
Figs. 9.4 (burst), 9.6 (periodic), and 9.7 (stochastic) along with the
sensitivities of present and proposed detectors.
The detectors in Section 9.5 are divided into those that operate in the
high-frequency regime, f <; 10 Hz (Sections 9.5.2, 9.5.3, and 9.5.4), those for
low frequencies, 10 Hz <; f <; 10 - 5 Hz (Section 9.5.5), and those for very low
frequencies, f ;S 10- 5 Hz (Section 9.5.6). The high-frequency detectors are
all earth-based; but because of seismic and gravity-gradient noise, the low
and very-low-frequency detectors must be space-based. Sections 9.5.2 and
9.5.3 describe in great detail the earth-based, high-frequency bar and beam
detectors which have been under development for many years and show
great promise for the future. Section 9. 5.4 describes briefly other types of
earth-based, high-frequency detectors. Section 9.5.5 describes low
frequency detectors including doppler tracking of spacecraft (subsection a),
beam detectors in space which hold great promise for the tum of the century
(subsection b ), the normal modes of the earth and sun (subsections c and d),
the vibrations of blocks of the earth's crust (subsection e), and the earth
orbiting skyhook (subsection f). Section 9.5.6 describes very-low-frequency
detectors including the timing of pulsars (neutron-star rotations), which
recently has placed interesting observational limits on a stochastic
338 K. S. Thorne
background (subsection a), the timing of orbital motions of binary and

planetary systems (subsection b ), and astronomical observations of
anisotropies in the temperature of the cosmic microwave radiation
(subsection c), deviations from the Hubble flow (subsection d), and products
of primordial nucleosynthesis (subsection d).
9 .1 .4 Notation and conventions
Throughout this chapter, unless otherwise stated, we shall assume that
general relativity correctly describes classical gravitational waves. Our
notation will be that of Misner, Thorne, and Wheeler ( 1973 ) - cited as MTW
throughout - including, e.g., the use of Greek indices for spacetime (running
from 0 to 3) and Latin for space (running from 1 to 3), the use of commas for
partial derivatives and semicolons for covariant derivatives, the use of the
Einstein summation convention, and the use of geometrized units in which
Newton's gravitation constant G and the speed of light c are set equal to
unity. Sometimes, particularly when discussing gravitational-wave
detectors, we shall restore the Gs and cs to the equations and use cgs units.
9.2 The physical and mathematical description of a gravitational
wave
9 . 2 .1 Shortwave approximation
General relativistic gravitational waves are ripples in the curvature of
spacetime that propagate with the speed of light. Because gravity is non
linear, it is not possible in a fully precise manner to separate the
contributions of gravitational waves to the curvature from the contributions
of the earth, the sun, the galaxy, or anything else; and since such a
separation underlies the very concept of a gravitational wave, this means
that gravitational waves are not precisely defined entities.
On the other hand, in realistic astrophysical situations the lengthscale ..t
on which the waves vary (their reduced wavelength, ..t = )./2n) is very short
compared to the lengthscales !l7 on which all other important curvatures
vary ; and this difference in lengthscale makes possible a high-accuracy, but
approximate, split of the Riemann curvature tensor Ra.fJylJ into a
'background curvature' R�fJylJ plus a contribution R��{J due to gravitational
waves : the background R�fJylJ is the average of Ra.fJylJ over several wavelengths
R�fJy{J = (Ra.fJy{J ) ; ( la)
and the waves' curvature R�p�lJ is the rapidly varying difference
R�p�{J = Ra.fJy{J - R�fJy{J · ( lb)
Heuristically, R�pyc> is like the large-scale ( 10 cm) curvature of an orange,

while R�p�c> is like the fine-scale granulation of the skin of the orange
(lengthscale a few millimeters).
This method of defining a gravitational wave, introduced into general
relativity by Wheeler ( 1955) and Power and Wheeler ( 1957), is a special case
of a standard technique in mathematical physics variously called 'shortwave
approximation' or 'two-timing' or 'two-lengthscale expansion' or 'two
variable expansion' (see e.g. Chapter 3 of Cole, 1968) ; and it is intimately
connected to the 'WKB approximation'. There is an elegant shortwave
approximation formalism for gravitational-wave theory due largely to Brill
and Hartle (1964) and to Isaacson ( 1968a,b). Not surprisingly the formalism
reveals that general relativistic gravitational waves propagate through
vacuum in essentially the same manner as light - with the same speed, with
the same changes of amplitude due to curvature of the wavefronts, with the
same diffraction effects when focussed by a gravitational lens, etc. Because
that formalism has been reviewed extensively elsewhere (e.g. Thome, 1983),
I shall not delve into it here, except for a brief description in Section 9.3.4
and an enumeration and brief discussion of some of its predictions in Section
9.3.5 below.
9 .2 .2 Measurements in the proper reference frame of an observer
In general relativity the Riemann curvature tensor is defined, operationally,
by the relative accelerations it produces between adjacent particles (e.g. the
'equation of geodesic deviation', Sections 8.7 and 1 1 .3 of MTW).
Correspondingly, a gravitational wave can be defined operationally in the
following way :
Consider an observer (freely falling or accelerated, it doesn't matter so
long as the acceleration is slowly varying). Let the observer carry with
herself a small Cartesian latticework of measuring rods and synchronized
clocks (a 'proper reference frame' in the sense of Section 13.6 of MTW, with
i
spatial coordinates x that measure proper distance along orthogonal axes).
She is to measure the 'force of gravity' Fi that acts on a particle of mass m,
i
momentarily at rest at location x . For example, she might let the particle
fall freely, measure its acceleration gi in her proper reference frame, and
multiply it by the particle's mass m to get the force Fi = mgi. Alternatively,
she might measure the force required to hold the particle fixed in the
coordinate grid of her proper reference frame and equate Fi to minus that
force. Of course, this is not different in any way from what she would do were
she a Newtonian physicist rather than a relativistic physicist. The difference
340 K. S. Thorne
lies in how she analyzes and thinks about the force of gravity Fi . As a
relativist, she recognizes (cf. Box 37. 1 of MTW) that Fi is made up of a nearly
steady, nearly position-independent component caused by her own failure
to fall freely, plus a component proportional to the particle's Cartesian
coordinate position x i (relative acceleration of particle and origin of
coordinates) which is caused by spacetime curvature R!%py� · This latter
contribution,
Fi = - mRjok o x k (2)
(where the index 0 denotes a component along her time basis vector), she
splits up into a piece that changes slowly in time (background curvature
contribution) plus a piece that is rapidly varying. She makes certain that
there are no rapidly moving or rapidly changing nearby sources of gravity to
account for the rapid variations; if there are none, then she can attribute the
rapidly varying component of the force to gravitational waves
FiGW -
_ - mR iGOkWo X k · (3)
It is conventional to use, as the primary entity for describing a
gravitational wave, not the Riemann curvature tensor R�� which has
dimensions 1/time 2 (or ljlength2), but rather a dimensionless 'gravitational
wave field' hf,[. In terms of the force-producing components of the Riemann
tensor R GwiOk o and proper time t as measured by our observer, h],[ is defined
by (cf. Section 2.3 of Thome 1983)
8 2 hf,!'_
8f = - 2R jGoWk o · (4)
The convenience of this gravitational-wave field lies in its simple

relationship to displacements produced by the waves : if, for simplicity, the
observer is freely falling and keeps the axes of her coordinate grid tied to
gyroscopes and is in a region of spacetime where gravitational waves are the
only source of spacetime curvature, then the waves will produce tiny
oscillatory changes bx i in the position of a test particle relative to the origin
of her coordinate grid ; and these changes will satisfy the equation of motion
('equation of geodesic deviation')
d2 bxj _ _1 m 82 h jTTk x k .
GW x k -
m - FjGW -
- - mR jOkO (5)
dt2 2 8t2
Because any realistic wave is so weak that the oscillatory changes bxi are
miniscule compared to the distance of the particle from the origin, x k can be
regarded as essentially constant on the right-hand side, and equation (5) can
then be integrated easily to give

bxi = � hTTxk
k • (6)
2 J
Thus, aside from a factor -i, h ]! plays the role of the 'dimensionless strain of
space', or the 'time-integrated shear of space' : it is the ratio of the wave
induced displacement of a free particle relative to the origin, to its orginal
displacement from the origin. (In the above equations and below it does not
matter whether a spatial index is up or down, since the spatial coordinates
are Cartesian.)
The superscript TT on the gravitational-wave field is to remind us that,
according to general relativity, the field is 'transverse and traceless'. More
specifically : the Einstein field equations guarantee that R�p�6 and hence also
h ],!' propagate with the speed of light. Since the sources of cosmic gravity
waves are very far away, the waves look very nearly planar as they pass
through the observer's proper reference frame. If we orient the x, y, z spatial
axes so the waves propagate in the z direction, then the 'transversality' of the
waves means that the only non-zero components of the wave field are h�,
h� = h�, and h-;:; and the 'trace-free' property means that h"IJ = - h"JJ .
Thus, the gravitational waves, like electromagnetic waves, have only two
independen� comp �nents - two polarization states.
Because of their TT nature, gravitational waves produce a quadrupolar,
divergence-free force field (equation (5) and Fig. 9. 1). This force field has two
components corresponding to the two polarization states of the waves : the
Fig . 9. 1 . Lines of force for gravitational waves (equations (5) and (7)) : (a )
with ' + ' polarization, m ()x = Mi+x and m ()ji = -Mi+ y; and (b) with ' x '
polarization, m ()x = 1/1 x y and m ()ji = -tli x x - where dots denote time
derivatives.
y y
(a } ( b}
342 K . S. Thorne
quantity
+h = xx -
- hIT - - hITyy (7a)
produces a force field with the orientation of a ' + ' sign, while
h =
x xy - h TT
- h IT yx (7b)
produces one with the orientation of a ' x ' sign. Thus, h and h are called
+
x
the 'plus' and 'cross' (or + and x ) gravity-wave amplitudes. From these
amplitudes and the polarization tensors e;x = - e� = 1, e;Y = e;x = 1 (all
other components zero), one can reconstruct the full wave field
h Jl = h + (jk + h " ei� . (7c)
It is straightforward to show that, if one rotates the x and y axes in the
transverse plane through an angle dl/J, the gravity-wave amplitudes are
changed to
h".;w h �d cos 2dl/J + h 0Jd sin 2dl/J,
=
h n,,ew = - h �d sin 2d l/f + h 0Jd cos 2dl/J. (7d)

The quadrupolar symmetry of the force field, together with the tenets of
canonical field theory, tells us that general relativistic gravitational waves
must be associated with quanta of spin two ('gravitons'). The spin is always
the ratio of 360 degrees to the angle, about the propagation direction,
through which one must rotate an instantaneous field to make it return to its
original state - the 'return angle'. For electromagnetic waves the return
angle is 360 degrees and the spin is one; for general relativistic gravitational
waves the return angle is 1 80 degrees and the spin is two. In other relativistic
theories of gravity the wave field has other symmetries and therefore other
spins for its quanta - and in most relativistic theories, by contrast with
general relativity, the symmetries are not even frame-invariant (local
Lorentz-invariant); and, as a result, the waves of those theories cannot be
incorporated into canonical field theory and cannot be quantized by
canonical techniques. For details see Eardley, Lee and Lightman ( 1973 ), and
Eardley et al. ( 1973).
The above definition of the gravitational-wave field h}J relies on a specific
choice of reference frame. It turns out that, if one pursues this definition in
two different reference frames related by a boost in some arbitrary direction,
and if one chooses the spatial axes of the two frames so they are unrotated
relative to each other, then the instantaneous wave fields h Jl (and
correspondingly h + and h " ) will be the same. Stated more precisely - but in
a language I shall not explain - the general-relativistic gravitational-wave
field h ft[ has 'boost-weight zero' (it is a scalar under boosts); but, as
discussed above, it has 'spin-weight two' (it behaves like a spin-two field
under rotations). For further details see Section 2.3.2 of Thorne ( 1983), and
for the mathematics that goes along with the concepts of 'spin-weight' and
'boost-weight', see Geroch, Held and Penrose ( 1973).
9.2 .3 TT coordinate system

The above description of the force law and force fields produced by a
gravitational wave is just the leading order in a power series expansion in
distance r = (c5 ikxix k yt from the origin of the observer's proper reference
frame. The higher-order fractional corrections to the forces are of order
(r/J..) 2 and, correspondingly, the proper reference frame's coordinates xi fail
to measure proper distance by fractional amounts of order hf,[ (r/J..) 2 (see,
e.g. Zhang , 1986, for discussion in the case of a freely falling observer). Thus,
the above description of gravity-wave forces is accurate only if the region of
interest is spatially small compared to a reduced wavelength. When the
region is large, an alternative description is needed.
The nicest alternative description makes use of a 'TT coordinate system'
(Section 35.4 of MTW), i.e. a coordinate system which is nearly Minkowski
throughout the spacetime region of interest, and in which the contribution
of the waves to the deviations from the Minkowski metric is embodied in the
same h j,! as we introduced above.
Because the TT coordinates must be nearly Minkowski, they cannot
cover a spacetime region that is too large : their extent in both time and space
must be far smaller than the background radius of curvature 9f8 - IR�p yb1 -t.
In typical situations, this limitation is very mild compared to the limitation
on the size of an observer's proper reference frame : one can stretch TT
coordinates over any region small compared to the Hubble distance (cutting
out holes in the vicinities of black holes and neutron stars), but if the waves
have a frequency of a kilohertz, then any proper reference frame used to
study their forces must be small compared to J.. - 50 km.
In a TT coordinate system the spacetime metric coefficients take the form
gap = 1'/a.p + h�p + h�, (8)
where 1'/a.p are the Minkowski metric coefficients (diagonal - 1 , + 1 , + 1 ,
+ 1), h�p i s the background metric perturbation which varies on a long
lengthscale !£, and h� is the gravity-wave metric perturbation which varies
on the short lengthscale .t The time-time and space-time components of the
.
gravity-wave perturbation h1J, and hlJ = h ]6 vanish, and the space-space

components hj,! are the same as the gravity-wave field that would be
344 K . S. Thorne
computed from the Riemann tensor (equation (4)) in the proper reference
frame of any observer who is nearly at rest in the TT coordinate system. For
proofs and discussions, see e.g. Sections 35.4, 37. 1 , and 37:2 of MTW.
Whereas physics in a proper reference frame can be formulated in the
Newtonian language of three-dimensional forces, including gravitational
forces, physics in a TT coordinate system must be formulated in the
relativistic language of geodesic motion and vanishing divergence of the
stress-energy tensor; cf. Section 9.5. 1 below.
9.2.4 Energy, momentum, and quantization of gravitational waves

In the 1940s and early 1950s a controversy raged over whether or not
gravitational waves can carry energy. The controversy was ultimately
resolved by Herman Bondi ( 1957) using a simple thought experiment : place
several beads on a rough stick and let a gravitational wave pass. The above
description of the forces produced by the waves (which was first understood
fully by Bondi and by Felix Pirani 1956 and their colleagues) guarantees that
- if the stick is not too rough - the wave will push the beads back and forth
on the stick, heating it. Surely, if the wave can heat a stick, it must carry
energy.
It was not until the late 1960s that a fully satisfactory mathematical
description of the energy in a gravity wave was devised : Richard Isaacson
( 1968a,b), using the shortwave approximation which he had developed in
detail on the basis of cruder earlier work of Wheeler ( 1955), Power and
Wheeler ( 1957) and Brill and Hartle ( 1964), introduced a stress-energy
tensor T�w for gravitational waves. This stress-energy tensor, like the
background curvature R!p.,,6 , is smooth on the lengthscale ,t; it is obtained, in
fact, by averaging the squared gradient of the wave field over several
wavelengths :
�
Ta�W = t � ( h L:a hilp ) ,
3 7 I. )
(9)
where ( - · - ) means 'average over several wavelengths'. (For a pedagogical
derivation and discussion see Sections 35. 7-35. 15 of MTW ; for a beautiful
rederivation by the method of averaged Lagrangians see MacCallum and
Taub, 1973.) If the waves are propagating in the z direction, this stress
energy tensor takes the standard form for a bundle of zero-rest-mass
particles (gravitons) moving at the speed of light in the z direction :
1
T�0w = - Tgw = - Tz�w = Tz�w = (8h +/8t) 2 + (8h x /8t) 2 ) . ( 10)
16n (
In order of magnitude, restoring the factors of G and c, the energy flux in

the waves if they have frequency f = c/2n);. is
_ yo w '.:::::'. n c3
f ( )
2(h2+ + h 2 ) 320 erg f 2 / h� + h :_ ) (1 1)
cm 2 s l kH z \ (10 - 2 1 )2
=
oz
4G x
·
The numbers in this equation correspond to a strongly emitting supernova

in the Virgo cluster of galaxies, where there are several supernovae per year.
-
Contrast this huge gravity-wave energy flux with the peak electromagnetic
flux at the height of the supernova, ,...,., 10- 9 erg cm 2 s - 1 ; but note that the
gravity waves should last for only a few milliseconds, while the strong
electromagnetic output lasts for days.
Corresponding to the huge energy flux (1 1) in an astrophysically
interesting gravitational wave is a huge occupation number for the quantum
states of the gravitational-wave field : it is not hard to show that for the
above supernova burst only a handful of quantum states are occupied ; and
they each contain n - 1075 gravitons (equations (6)--(8) of Thome et al.,
1979). This means that the waves behave exceedingly classically; quantum
mechanical corrections to the classical theory have fractional magnitude
1/Jn ,...,., 10- 37. (Although the full quantization of the gravitational field is
exceedingly difficult and not yet fully under control, the quantization of
weak gravitational waves propagating through a smooth background
spacetime - .equivalent to weak waves in flat spacetime - has been well
understood for decades; see, e.g., the most elementary aspects of Feynman,
1963; Dewitt, 1967a,b.)
Isaacson's stress-energy tensor (9) for gravitational wavr-s has the same
properties and plays the same role as the stress-energy tensor for any other
field or form of matter in the background spacetime. For example, Ta�w
generates background curvature through the Einstein field equations
(averaged over several wavelengths of the waves); also T;/iw has vanishing
divergence (conservation of gravity-wave energy and momentum) in
spacetime regions where the waves are not being generated, absorbed, or
scattered. For full details see Isaacson (1968b) or Section 35.15 of MTW.
9.3 The generation and propagation of gravitational waves
9.3 . 1 Wave propagation split off from wave generation
Turn, now, to the generation of gravitational waves and their propagation
from their source to the earth. Mathematically, the wave-generation
problem and the wave-propagation problem are each difficult - though for
very different reasons, so that to handle the difficulties requires two very
346 K. S. Thorne
different sets of mathematical tools and approximations. For ease of

analysis, then, it is important to split the propagation problem off from the
generation problem and treat them separately. This is accomplished by
dividing the space around the source into three regions (Section III of
Thorne, 1980b) : a 'wave-generation region' at distances from the source
r� r1 = ('Inner radius'); a 'local wave wne' at distances r1 � r � r0 = ('Outer
radius'); and a 'distant wave zone' at distances r ;z: r0. The theory of wave
generation is developed with one set of mathematical tools in the wave
generation region and the local wave zone, i .e. at distances r � r0; the theory
of propagation to earth is developed with the other set of tools in the local
wave zone and the distant wave zone, i.e. at distances r ;z: r1 ; and the two
theories are matched together in their domain of overlap, the local wave
zone r1 � r � r0 •
The inner radius r1 is far enough out to be in the wave zone, r1 � ..t; far
enough to be in a region where the source's gravity is weak, r1 � 2M =
(Schwarzschild radius of source) = 2 x (mass of source) ; and far enough to be
outside the source, r1 � L = (size of source). The outer radius r0 is far enough
beyond the inner radius to leave many wavelengths in the local wave zone,
r0 - r1 � ..t ; but not so far that the gravitational redshift can produce a
significant net phase shift during propagation through the local wave zone,
·
b</> = (M/..t)ln(r0/r1 ) � 1 ; and not so far that the background curvature of the
external universe can significantly affect the propagation, r0 - r1 � 9f 8 =
IR!pyal -! = (radius of curvature of background spacetime). These choices of
r1 and r0 permit one to ignore, in the local wave zone r1 � r � r0, the
background curvature both of the source and of the external universe ; i .e.
they permit one to regard the waves in the local wave zone as propagating
through flat spacetime. This greatly simplifies calculations.
For a given, astrophysically interesting source the wave-generation task
consists of computing with reasonable accuracy the dynamical behavior of
the source's gravitational field in the wave-generation region r� r1, and
further computing how that dynamical curvature develops into outward
propagating waves in the local wave zone, r1 � r � r0. Once this is done, the
theorist can switch tasks and mathematical formalisms from wave
generation to wave propagation. The wave-propagation task takes as input
the propagating waves in the local wave zone and carries them outward
(typically using the shortw<!.ve approximation and formalism) through the
universe from source to earth .
As an addendum to the wave-generation task , one often computes the
back-reaction effects of the wave emission on the source, i.e. the 'radiation
reaction'.
9 .3 .2 The quadrupole formalism for wave generation and radiation

reaction
Of all the techniques for computing wave generation, one, the 'quadrupole
formalism', is especially important because it is highly accurate for many
sources and is accurate in order of magnitude for most. The quadrupole
formalism was derived originally by Einstein ( 19 16, 19 18) for sources with
negligible self-gravity and slow internal motions. Later, in a series of steps by
Landau and Lifshitz ( 194 1), Fock ( 1959), lpser ( 1 97 1) and Thome ( 1980b,
Sections VII and XII), it became clear that the quadrupole formalism
requires for high accuracy no constraints whatsoever on the strength of the
source's internal gravity; all that is required is slow motion - more
specifically, that the source's size L be small compared to the reduced
wavelength X of the waves it emits. In other words, the quadrupole
formalism is to gravity-wave generation what the 'poor-antenna
approximation' (dipole formalism) is to radio-wave generation. Moreover,
as in the radio-wave problem, the quadrupole formalism typically is
accurate to within factors of order 2 even for sources with sizes of order a
reduced wavelength X (cf. equation (4.8) of Thome, 1980b); and since very
few astrophysical systems are larger than a reduced wavelength of the waves
they emit, this justifies the use of the quadrupole formalism for order-of
magnitude astrophysical estimates in almost all situations.
The quadrupole formalism writes the gravitational-wave field in the
source's local wave zone (where the background curvature can be ignored)
in the following simple form :
TT
= - ;i0 [fjk ( t - r)]TT .
2 2
hik ( 1 2)
r ut 2
Here r is the distance to the source's center, t is proper time as measured by
an observer at rest with respect to the source, t - r is retarded time, the
superscript TT means 'algebraically project out and keep only the part that
is transverse to the (radial) direction of propagation and is traceless' (Box
35. 1 of MTW), and efik (t - r) is the source's mass quadrupole moment
evaluated at the retarded time t - r.
The meaning of 'mass quadrupole moment' is well known when the
source has weak internal gravity and small internal stresses, so Newtonian
348 K. S. Thorne
gravity is a good approximation to general relativity inside and near the

source. Then fik is the symmetric, trace-free (STF) part of the second
moment of the source's mass density p , as computed in. a Cartesian
coordinate system centered on the source :
/;,(t) = [J
p(t)x i x• d3x JTF J { � J
= p(t x ix• - r2 b; d 3x ;
• ( 1 3a)
equivalently, it is the coefficient of the l/r3 part of the source's Newtonian
gravitational potential
k
M 3 fjk( t)xix k 5 Aik x i x ix . . .
 = -- -- --
+ ( 1 3b)
r 2 r5 2 r7 •
If the source has strong internal gravity, one can no longer express its
mass quadrupole moment as the simple integral ( 1 3a). However, so long as
the source has slow internal motions (L � .t), there will be a region of space
far enough from the source to be in vacuum (r > L) and far enough for
gravity to be weak (r � 2M), yet near enough for retardation and wave
behavior to be unimportant (r � .t). In this 'weak-field, vacuum, near-zone'
gravity can be described with high accuracy as Newtonian ; and the
Newtonian potential is "" -�(g00 + 1), where g00 is the time-time part of
the metric in a coordinate system that is as Minkowski as possible
throughout the weak-field, vacuum near zone. The mass quadrupole
moment can then be read off this Newtonian potential using the standard
formula (13b). For further discussion see the review in Section 3 of Thorne
(1983) or the original treatment in Part Two of Thorne ( 1980b).
For order-of-magnitude calculations of hf,! (equation (12)), one can
approximate the TT part of the second time derivative of the mass
quadrupole moment by that portion of the source's internal kinetic energy
which is associated with non-spherical motions, Ekrn ; and, unless the source
is at very large cosmological redshifts z � 1 , one can propagate the waves to
earth as though the intervening spacetime were flat - with the result that the
local wave-zone formula for the waves, equation ( 12), is valid also at earth.
The result is the simple order-of-magnitude formula
( 14)
for the magnitude h of the gravity-wave field hf,! at earth.
From the 'exact' quadrupole formula ( 12) for the wave field in the local
wave zone and Isaacson's formula (9) for the stress-energy tensor of the
waves, one can compute the fluxes of energy and angular momentum carried
by the waves. By integrating those fluxes over a sphere surrounding the

source in the local wave zone, one obtains for the rates of emission of energy
(Einstein, 1916, 1918) and angular momentum (Peters, 1964)
dEGW �
dt
= (( ) )
5 j?:, k
d3 fjk 2
dt3 '
(15)
dt
=
5 j.L.
( )
dJ y w � " f> · 8 2 fj1 8 3 "'k
k
,k,l '1 8 t 2 8 t3 •
( 16)
The linear momentum carried off by the waves vanishes when one computes
it by the quadrupole formalism ; but when one includes higher-order
corrections to the field emitted by slow-motion sources, one finds for the rate
of emission of linear momentum (first derived by Papapetrou, 1962, 197 1 ;
for a more modem derivation in the notation of this chapter see Section
IV.C of Thorne, 1980b)
dt
= (
dP yw 2_ "'°' 8 3 ,fik 8 4 ,/jki
� 8 t3 8 t4
63 J,k
) + .L. (
45 J,k,a ljk 8 t3 8 t3
)
16 "'°' t;. . 8 3 /ja 8 3 �a .
(l7)
Here Jiik is the source's 'mass octupole moment' and !Iii is its 'current
quadrupole moment' (gravitational analog of magnetic quadrupole
moment). For sources with weak internal gravity and stresses (nearly
Newtonian sources), these moments are computable from the simple
volume integrals
i i STF
fiik = (J p x x x k d3x ) , ( 18a)
!/:I). = (J p t; .lp xPv q xi d 3 x ) STF ' ( 18b)
q
where, as in equation (15a), STF means 'make it symmetric and trace-free',
i .e. 'symmetrize on all free indices and remove the traces on all pairs of free
indices'. Independently of the strengths of the internal gravity and stresses,
the moments can be read off the Newtonian potential <l> ""' - 1(g 00 + 1)
(equation ( 1 3b)) and off the 'gravitomagnetic potential' {3i - 90i in the
source's weak-field near zone :
( 19)
In equation ( 19) J P, the moment in the leading, dipolar term, is the source's
angular momentum. For further details, discussions, and derivations see
Thorne ( 1983) or Thome ( 1980b). For a discussion of the gravitomagnetic
potential see, e.g., Chapter 3 of Thome, Price and Macdonald (1986).
The laws of conservation of energy, angular momentum, and linear
momentum imply that radiation reaction should deplete the source's
350 K . S. Thorne
energy, angular momentum, and linear momentum at just the right rates to
compensate for the losses ( 15), ( 16), and ( 17); and a detailed analysis of the
radiation reaction forces reveals that this is so (Peres, 1960). Particularly
convenient in analyzing the radiation reaction in a source with weak self
gravity is a Newtonian-type radiation-reaction potential (Burke, 1969 ;
Thorne, 1969; Chandrasekhar and Esposito, 1970; Section 36.8 of MTW).
Different physicists feel comfortable with different levels of rigor. In recent
years these differences have shown up strongly and publicly in a controversy
over derivations of the quadrupole wave-generation formula ( 12) and the
formula for the energy sapped from a source by radiation reaction (negative
of equation ( 15)). Many physicists - myself among them - were quite
satisfied with derivations at the level of rigor, e.g. , of Landau and Lifshitz
( 194 1) and Peres ( 1960). Others (e.g. Ehlers et al., 1976) felt that those early
derivations were inadequately rigorous and, correspondingly, that the
quadrupole formulae were suspect for sources with non-negligible self
gravity. The controversy was heightened by the fact that there were
mathematical errors in some (but not all) of the early derivations (see
Walker and Will, 1980 and Section 3 .4.2 of Thorne, 1983 for discussions).
The controversy was still raging in the early 1980s; see, e.g. Ashtekar
( 1983). However, duiing the mid 1980s it has largely subsided. There are
now many new derivations of the quadrupole formulae, with much
improved rigor, and they all produce the same, standard results; see, e.g.,
Anderson et al. ( 1982), Blanchet and Damour ( 1984) ; Christodoulou and
Schmidt ( 1979); Isaacson, Welling and Winicour ( 1984) ; and for reviews see
Will ( 1986), Schutz ( 1986a) and Damour ( 1987).
Of particular interest is radiation reaction in the binary pulsar
PSR 19 13 + 16, which should cause the binary system's two neutron stars to
spiral together slowly with a consequent gradual decrease in their orbital
period. Because the duration of observations of the pulsar is short ( 12
years), the cumulative effects of radiation reaction on the orbit during those
observations are 100 times smaller than post-Newtonian effects; and,
consequently, the detailed effects of the radiation reaction could not be fully
understood until the orbital equations were fully under control up through
post-post-Newtonian order. Damour and Deruelle ( 1986) have now
brought the orbital equations fully under control ; and there is now a
beautiful agreement between those equations - including the quadrupolar
radiation reaction - and the observational data. For a detailed discussion
see Chapter 6 of this book.
9.3 .3 A catalog of formalisms for computing wave generation

The order-of-magnitude formula h "" Ekf /r shows that of all sources at fixed
n
distance r, the strongest emitters will be those with the largest non-spherical
kinetic energies, i.e. those with the largest internal masses and with internal
velocities approaching the speed of light. Thus, the strongest emitters are
likely to violate the slow-motion assumption which underlies the
quadrupole formalism , and will require for accurate analysis either higher
order corrections to the quadrupole formalism, or wave-generation
formalisms that do not entail any slow-motion assumption.
There are a number of other wave-generation formalisms which can be
applied to such sources. This section is a catalog of them , with references to
detailed presentations and applications. For an out-of-date but unified
presentation of most of these formalisms see Thorne ( 1977).
To be tractable with a minimum of numerical computation, a wave
generation formalism must break the extreme non-linearity of the Einstein
field equations by imposing a power-series expansion in some small quantity
and keeping only the lowest , linear order or the lowest few orders. Wave
generation formalisms can be classified according to their choice of the small
expansion parameter. Slow-motion formalisms (of which the quadrupole
formalism is an example) expand in L/Jc = (size of source)/(reduced
wavelength of waves) ; subsection (a) below. Post-Minkowski formalisms
expand in the strength of the gravitational field inside the source, i.e. in the
magnitude of the deviations of the metric coefficients from their Minkowski
values ; subsection (b ). Post-Newtonian formalisms expand simultaneously
in L/Jc and the strength of the internal gravitational field; subsection (c).
Perturbation formalisms expand in the deviations of the metric from its form
for some non-radiative, astrophysical system - e.g. from the Kerr metric for
a rotating black hole, or from the metric for an equilibrium, rotating,
relativistic stellar model, or from the Friedman-Robertson-Walker metric
for a homogeneous, isotropic 'big-bang' ; subsection (d).
Of all astrophysical sources, the very strongest emitters will entail
gravitationally induced large-amplitude, high-velocity, non-spherical
internal motions - e.g. the inspiral and coalescence of a binary black hole or
binary neutron-star system . For such sources there is no small parameter in
which one can expand . The only way to compute the full details of the wave
field emitted by such sources is by the techniques of numerical relativity : the
numerical solution of the full Einstein field equations on a supercomputer;
subsection (e).
3 52 K . S. Thorne
(a) Slow-motion formalisms

As for electromagnetic waves, so also for gravitational waves, slow-motion
expansions give rise automatically to multipolar expansions : the
electromagnetic vector potential is a sum of an electric di polar term (typical
magnitude (Q/r)(L/J.) where Q is the charge of the source), plus an electric
quadrupole and a magnetic dipole [magnitude (Q/r)(L/J.) 2] , plus an electric
octupole and a magnetic quadrupole [magnitude (Q/r)(L/J.)3] , etc.
Similarly, the gravitational-wave field hf,[ is the sum of a mass-quadrupole
term [ - (8 2/8t 2),fik /r - (M/r)(L/J.) 2] , plus a mass octupole
[ - (83/8t3)/iik /r - (M/r)(L/J.)3] and a current quadrupole
[ - (8 2/8t 2 )f/ii /r - (M/r)(L/J.)3] , etc. In each case, electromagnetic and
gravitational , there are two families of moments involved ; and in each case
as one goes to higher orders in L/J. one is driven to include higher-order
moments of the source.
Early foundations for these expansions in the gravitational case will be
found in Bonnor ( 1959) and Pirani ( 1964) ; full mathematical details will be
found in Thorne ( 1980b) ; and major improvements and elucidations will be
found in Blanchet and Damour ( 1986), Blanchet ( 1987a) and Damour
( 1987).
A good example which shows how, as the source's internal motions are
speeded up, the higher moments gradually become more and more
important, is the gravitational bremsstrahlung radiation emitted when two
stars fly past each other with some high initial velocity ; see Kovacs and
Thorne ( 1978) and Turner and Will ( 1978) for full details.
In some slow-motion systems the mass quadrupole contribution may be
suppressed, leaving the current quadrupole or the mass octupole to
dominate. A good example is the torsional oscillation of a neutron star, in
which the motions are slow because the shear modulus is weak, and the mass
moments vanish because of parity considerations leaving the current
quadrupole to dominate the radiation ; for details see Schumaker and
Thorne ( 1983). Another example is the Chandrasekhar ( 1970)-Friedman
Schutz ( 1978) (CFS) instability in neutron stars, which preferentially excites
mass octupole (l = 3) or hexadecapole (l = 4) or l = 5 modes of pulsation
causing them to radiate more strongly than quadrupole; see Section 4.3(b)
below.
(b) Post-Min k owski formalisms
Post-Minkowski wave-generation formalisms are sometimes called 'post
linear' because they entail expanding in the strength of the gravitational field
beyond linear order; and they are sometimes called 'fast-motion' to contrast
them with slow-motion formalisms.
There is a systematic way to take a post-Minkowski wave-generation
formalism that is accurate to a given order in the strength of the source's
internal gravity, and iterate it to obtain a formalism of higher accuracy
(Thorne and Kovacs , 1975 ; Thorne, 1977).
The wave-generation formalism that is accurate to first post-Minkowski
order (first order in the strength of internal gravity) is 'Linearized theory',
i .e. the linear approximation to general relativity. Linearized theory is
discussed in most textbooks, e.g. Chapter 18 and Sections 35. 1-35.6 of
MTW . Halpern and Desbrandes ( 1969) and, independently, Press (1977)
have derived a particularly useful Linearized wave-generation formula for
systems with sizes L large compared to a reduced wavelength ..t. Examples of
gravity-wave generation that have been analyzed by Linearized theory are :
(i) the coherent (but painfully slow) transformation of electromagnetic
waves into gravitational waves (first considered by Gertsenshtein , 1962,
subsequent work reviewed in Section 4. 1 of Grishchuk and Polnarev, 1980);
and (ii) the waves emitted by the explosion of a non-spherical nuclear bomb
(Wheeler, 1962; Wood et al., 1970).
Linearized theory is completely ignorant of the source's internal gravity ;
it can correctly predict the emitted waves only if the source's motions are
governed by non-gravitational forces - typically by electric or magnetic
forces. For systems with significant but weak internal gravity (e.g. stellar
pulsations, binary systems, and high-speed stellar encounters), one must use
a wave-generation formalism accurate to the next, 'post-post-Minkowski'
or 'post-Linear' order. For the details of such a formalism, see, e.g . , Thorne
and Kovacs ( 1975), Crowley and Thorne ( 1977); and for its application to
high-speed stellar encounters (gravitational bremsstrahlung radiation) see
Kovacs and Thorne ( 1977, 1 978). For a recent review see Westpfahl ( 1985).
Thus far nobody has developed a post 2 -Linear wave-generation
formalism in detail - i .e. a formalism accurate to post 3 -Minkowski order.
There has been no great need for such a formalism, and the post-Linear
formalism is sufficiently hard to work with in practice (cf. Kovacs and
Thorne, 1 977) that it is not clear whether post 2 -Linear would be
significantly easier than full-blown numerical relativity.
(c) Post-Newtonian formalisms

Post-Newtonian approximations to general relativity make the assumption
- in accord with the virial theorem for gravitationally bound systems - that
354 K . S. Thorne
inside the source the deviations of the metric from Minkowski (i.e. the
dimensionless strength of the source's gravity) have a magnitude e of order
(L/J.) 2 ; and accordingly they expand the Einstein field equations
simultaneously in e (post-Minkowski expansion) and L/J.. (slow-motion
expansion). See, e.g., Burke ( 1979) for a review of the method.
The lowest-order wave generation formalism that results from this
expansion is called 'Newtonian' because it computes the evolution of the
source using Newton's laws of gravity and mechanics, then evaluates the
source's time-evolving quadrupole moment using the standard Newtonian
volume integral ( 1 3a), then inserts that quadrupole moment into the
standard quadrupole wave-generation formula ( 12). It is this Newtonian
version of the quadrupole formalism that has been especially controversial
(see the end of Section 3 .2 above) but is now almost universally agreed to be
highly reliable.
There have been a large number of important wave-generation
calculations with this formalism. Some examples are : (i) the waves emitted
by binary systems in Newtonian, elliptical orbits (Peters and Mathews,
1963); (ii) the waves emitted by a variety of models of stars that collapse to
form neutron stars (Saenz and Shapiro, 1978, 198 1 ) ; and (iii) the waves
emitted in the .head-on' collision of two compact stars (Gilden and Shapiro,
1984).
The Newtonian wave-generation formalism starts losing accuracy when
the source's internal gravity becomes too strong {t: - 0.05) and its internal
velocities too high (v - 0.2) (Turner and Will, 1978) - e.g. in the late stages of
the spiraling together of a neutron-star binary system. In such a situation it
is useful to include the next higher-order corrections (one order higher in the
strength of gravity e , two higher in the speed L/J.. ) . The result is the post
Newtonian formalism (Epstein and Wagoner, 197 5 ; Wagoner, 1977 ;
Tsvetkov, 1984). Examples of calculations that have been performed with
the post-Newtonian formalism are the radiation from a system of bodies
whose sizes are all small compared to their separations (Wagoner and Will ,
1976), gravitational bremsstrahlung at moderate velocities (Turner and
Will , 1978), and the radiation emitted by a slowly rotating star that collapses
to a neutron star (Turner and Wagoner, 1979).
The foundations for a post 2 -Newtonian wave-generation formalism have
also been worked out (Section V . E. of Thorne, 1980b) ; but it has never been
developed in full detail . or applied to any sources. Such a formalism - by
contrast with the post 2 -Linear - would likely be far more tractable than full-
blown numerical relativity; so it may one day prove useful in tying down the
gravitational waves from large-amplitude processes involving neutron stars.
(d) Perturbation formalisms

Much can be learned about gravitational waves from neutron stars and
black holes by studying weak , non-radial perturbations around their
equilibrium structures. The theory of non-radial perturbations of non
rotating relativistic stars was developed by Campolattaro, Detweiler, lpser,
Price and Thorne (see pp . 195-20 1 of Thorne 1978 for a review and
references) ; and extensions of the theory have been given by Schumaker and
Thome ( 1983), Detweiler and Lindblom ( 1985) and Finn ( 1986). Among its
recent applications are computations by Lindblom and Detweiler ( 1983) of
the normal modes of neutron stars with a variety of equations of state. For
rotating stars the corresponding theory is due largely to Chandrasekhar,
Friedman and Schutz (reviewed through 1977 on pp. 20 1-8 ofThorne 1978
and reviewed more recently by Schutz 1987). The most important recent
applications are studies of the waves emitted by the 'Chandrasekhar
Friedman-Schutz instability' in rapidly rotating neutron stars ; see Section
9.4.2(b) below .
For non-rotating (Schwarzschild) black holes the perturbation theory is
due to Regge and Wheeler ( 1957) with maj or subsequent contributions by
Bardeen , Chandrasekhar, Detweiler, Edelstein, M oncrief, Press,
Vishveshwara and Zerilli ; see pp. 180-8 of Thorne ( 1978) for a review.
Recent applications include a definitive numerical evaluation of the
eigenfrequencies and damping times of a Schwarzschild hole's quasinormal
modes (Leaver, 1985, 1 986a), and of the Green's function for arbitrary
perturbations of a Schwarzschild hole (Leaver, 1986b). Finally, for rotating
(Kerr) black holes the theory i s due to Teukolsky ( 1972, 1973), Teukolsky
and Press ( 1974), Wald ( 1973), Chrzanowski ( 1975), Cohen and Kegeles
( 1975), Chandrasekhar and Detweiler ( 1976), Detweiler ( 1977),
Chandrasekhar ( 1983) and Sasaki and Nakamura ( 1982) ; and recent
applications include the evaluation of the eigenfrequencies and damping
times of a Kerr hole's quasinormal modes (Detweiler, 1980; Leaver, 1985,
1986a), and evaluations of the gravitational waves emitted when a compact
body orbits a Kerr hole (Detweiler, 1978), scatters gravitationally off a Kerr
hole (Kojima and Nakamura, 1984b), or plunges into a Kerr hole (Detweiler
and Szedenits, 1 979 ; Kojima and Nakamura, 1984a).
Primordial gravitational waves (waves created in or near the big-bang)
cannot be analyzed by splitting space into a wave-generation region plus
356 K. S. Thorne
wave zones (Section 9.3. 1), because the transition from wave generation to
wave propagation is a temporal rather than a spatial one : the transition
occurs when the size of the cosmological horizon expands to become much
larger than a wavelength, thereby unfreezing a set of frozen-in initial
perturbations. The only way, today , to analyze primordial waves is by a
perturbation formalism somewhat akin to that used for stars and black
holes : the unperturbed configuration is typically a non-radiative,
I
Friedman-Robertson-Walker cosmological model , and the perturbations

are studied by linearizing the Einstein equations - or a quantized variant of
them - around that model . The resulting theory is due to Lifshitz ( 1 946) ; see
also Section 7.3 of Zel'dovich and Novikov ( 1 983), and references cited in
Section 9.4.3(d) below.
(e) Numerical relativity

Numerical solution of the full Einstein equations is the only way, today, to
study wave generation in the strongest and most interesting of all gravity
wave sources : those with high internal velocities, strong internal gravity,
and large deviations from a non-radiating spacetime. During the past
decade several dozen researchers have worked vigorously to develop the
field of numerical relativity. The results of this effort are : an analytic
formulation of the initial value problem and the dynamical evolution
problem for the Einstein field equations, in a form that facilitates numerical
solutions (York, 1983 , and references cited therein); for axisymmetric
systems, viable ways to slice spacetime and choose the spatial coordinates so
as to avoid pathologies and to compute the emitted gravitational waves
efficiently (Smarr and York , 1 978; Pi ran, 1983 ; Bardeen and Piran, 1983 ;
Sasaki, 1 984; Stewart and Friedrich , 1983 ; Isaacson, Welling and Winicour,
1983 ; Gomez et al. , 1986; Anderson and Hobill, 1 986; Evans and
Abrahams, 1987); and several good computer codes for evolving
axisymmetric systems and computing their waves (Nakamura, 1983 ; Stark
and Piran , 1 986; Piran and Stark, 1986; Evans, 1986).
Among the problems that have been studied successfully with the codes
thus far are the gravitational radiation produced by a head-on collision of
two Schwarzschild black holes (Smarr, 1977a), by the collapse of a rotating
star to form a Kerr black hole (Stark and Piran, 1986), and by the vibrations
of a neutron star (Evans, 1986).
The experts in numerical relativity are now beginning to move on from
axisymmetric systems, which have two non-trivial spatial dimensions and
one non-trivial time, to asymmetric, generic systems with three non-trivial
spatial dimensions and one time (e.g. Nakamura, 1987). This full '3 + 1'
effort will require using the world's largest supercomputers, and will require
new techniques for slicing spacetime into space plus time, for choosing the
spatial coordinates, and for differencing the Einstein equations. The effort
may absorb almost as many person-years as the development of
gravitational-wave detectors ; but it will be well worthwhile : the payoffs will
include the ability to compute in detail the waveforms from the strongest
gravity-wave sources in the universe, such as the spiraling together and
coalescence of two black holes - waveforms that will be crucial to the
interpretation of gravity-wave observations and to their use for strong-field,
highly dynamical tests of general relativity.
One should not be misled into believing that numerical relativity will be
the totally dominant tool for realistic gravitational-wave calculations in the
coming years. On the contrary, we can expect a healthy interaction between
numerical and analytical techniques; for discussion see Schutz ( 1986c).
9.3.4 Wave propagation in the real universe

Since I have recently written a detailed review of the theory of gravitational
wave propagation (Section 2 of Thorne, 1983), I shall only sketch the main
points briefly.
No matter how strong a source of gravity waves may be, once its waves
are fully formed (once they have reached a location, e.g. in the local wave
zone, where the inhomogeneity lengthscale ff' of the background curvature
is large compared to their wavelength ..t), they will have dimensionless
amplitudes h small compared to unity. This can be seen as follows :
The stress-energy tensor of the gravitational waves, Tfpw (equation (9)),
acts as a source for the background curvature through the Einstein
equations G�p = 8n(Ta�w + I'a/Jther). Since the background Einstein tensor
G�p has magnitude less than or of order 1/&lij (where &le is the background
radius of curvature as defined from the Riemann tensor), and since the
magnitude of the gravity-wave stress-energy tensor (9) is h 2/l 2 , the Einstein
equations imply that
(20)
Since the inhomogeneity lengthscale ff' of the background curvature is
always less than or of order the radius of curvature &le, equation (20) implies
the claimed result :
h � l/Ble � l/.ff' � 1 . (2 1)
This permits one to study the subsequent propagation of the waves, once
358 K. S. Thorne
they are fully formed, using a linearized approximation to the Einstein field
equations (Sections 35. 13 and 35. 14 of MTW) : (i) one introduces a field hap
(which is actually the trace-reversed contribution of the waves to the
spacetime metric in a suitable gauge). This field is defined to be equal to h r/
in the region from which the waves are propagating (the local wave zone in
the case of isolated sources ; the very early universe in the case of primordial
waves). One then evolves the field hap out into the surrounding universe and
to earth using the curved-spacetime wave equation (equation (35.64) of
MTW)
hap J ) µ + g:p hµ vJ - 2 h ( ) µJ P > + 2R!a. p hµ v - 2R!(ahp)µ = - l6 m5 Ta.p ·
vµ µ v
(22)
Here g �P is the background metric, the subscript and superscript I denote
covariant derivatives with respect to the background metric, R�p and R�pyb
are the Ricci and Riemann curvature tensors of the background ; and bTap is
the perturbation in the non-gravitational stress-energy tensor produced by
the trace-reversed metric perturbation ha.p itself.
Although the field ha.p initially is wavelike and thus has ,t � !e ;$ al8, it
might propagate into regions where the background has very short
lengthscales, .fR ;$ l. (For example, the waves produced by the Crab pulsar,
with ,t- 1000 km may propagate through a massive white dwarf with
!e - 1000 km and al8 - 30 000 km, or even through a neutron star with
!e - 10 km and al8 - 30 km). If this happens, one need not worry. The wave
equation (22), because it depends for its validity only on the weakness of the
ap
field h and not on the shortwave assumption, remains valid and carries the
field through the region of short background lengthscale (where, strictly
speaking, it is no longer a gravitational wave), and thence onward into
regions of long lengthscale (where it once again is a gravitational wave).
In those regions where hap is a wave, i .e. has ,t � !e, one can compute from
it the gravitational-wave field hJkT by a very simple prescription : introduce
the proper reference frame of a specific observer; and in that frame discard
the time-time and time-space parts of hap' and algebraically project out
from the space-space parts those pieces that are transverse to the
propagation direction and are trace-free. The result will be hf,[. For further
discussion and justifications see Box 35. 1 of MTW and Section 2.4.2 of
Thorne ( 198 3).
As I shall discuss below, the effects of the wave-stimulated stress-energy
perturbations bTap are never large enough to be astrophysically important,
so one can ignore them in propagation calculations. Moreover, in regions
(nearly everywhere) where Jc � �B ' the contributions of the background

curvature tensors R �a v/J and R �a to the wave equation can be ignored , and
one can specialize the gauge so as to make hap divergence free. The wave
equation (22) then assumes the simplest form imaginable :
0 hap ha/Ji/ = 0. (23)
When , in addition, the radius of curvature of the wave fronts is large
compared to A (as is true everywhere except near the very rare focal points of
gravitational lenses), the wave equation (23) can be solved easily by the
techniques of geometric optics (Isaacson, 1968a ; Exercise 35. 15 of MTW ;
Section 2.5 of Thorne, 1983) : the field fia/J propagates along null rays ; its
polarization is parallel transported along the rays ; and its amplitude, like
the amplitude of light, varies along each ray as 1/(the radius of curvature of
the wave front) .
Because good gravitational lenses are so rare in the real universe, and
because regions (black holes and neutron stars) with very strong curvature
(�B ;:5 Jc) are so rare and so small , the waves from almost every source will
propagate to earth via pure geometric optics. Moreover, because the
universe is almost globally Lorentz (flat) in its background geometry on
lengthscales small compared to the Hubble distance, for sources at distances
much less than Hubble (at cosmological redshifts z � 1), this geometric
optics propagation will produce a simple 1/r falloff of amplitude and will
preserve the wave form (the dependence on retarded time) and the
polarization . More specifically , it will produce a gravitational-wave field in
TT coordinates with the simple form
(24)
where r is distance to the source and {), ¢ are spherical polar angles centered
on the source. For a slow-motion source the function A I[ is just twice the TT
part of the second time derivative of the source's quadrupole moment, as
one sees trivially by matching (24) onto ( 12) in the source's local wave zone.
For a source at a large cosmological redshift z � 1, if one approximates the
background spacetime geometry by that of a Friedmann-Robertson
Walker cosmological model , the geometric-optics propagation produces the
same effects for gravity waves as for light : (i) the magnitude of h JkT falls off
with the same 1/R behavior as for light, where R is ( 1/2rc) x (circumference of
a sphere passing through the earth and centered on the source, at the time
the waves reach earth) (equations (29.28)-(29.33) of MTW); (ii) the
polarization, like that of light in vacuum, is parallel transported radially
360 K. S. Thorne
from source to earth; and (iii) the time dependence of the wave form is
unchanged by propagation, except for a frequency-independent redshift
!received /!em itted = 1/(1 + z). For further details and derivations see Section
2.5.4 of Thome ( 1983) or Section 7.2 of Thome ( 1977).
9.3 . 5 A catalog of wave-propagation effects
In principle gravitational waves can experience almost all the familiar
peculiarities of propagation that electromagnetic waves experience. Here I
shall enumerate those that have been studied, mention briefly their
importance or unimportance, and give references for further detail .
(a) Absorption, scattering and dispersion by matter and electromagnetic fields

Gravitational waves are so weakly absorbed by matter (Section 2.4.3 of
Thome, 1983) that absorption is astrophysically important only near the
Planck era of the big-bang (Section 7.2 of Zel'dovich and Novikov, 1983).
For experimenters , however, the tiny absorption that should occur in a
gravity wave detector is very important; see Section 9.5.2(b) below.
The scattering of gravitational waves by matter is also so weak that it is
never astrophysically important, except near the Planck era. In the analysis
of resonant gravity wave detectors, however, scattering is conceptually
important : the method of detailed balance (an idealized calculation in which
a resonant detector is driven by monochromatic waves into vibrations of
such unrealistically high amplitude that it reradiates at the same rate as it
absorbs - i .e. it scatters the waves strongly) is a powerful way of computing
the cross-section of a resonant detector; see Section 37.7 of MTW.
Coherent scattering by a medium produces dispersion - i .e. frequency
dependent propagation speeds. Although dispersion is very important for
electromagnetic waves, it is never of any importance in the real universe for
gravitational waves (Section 2.4.3 of Thome, 1983). It is instructive,
however, to imagine and study theoretically an unrealistic form of matter
('respondium') with such strong dispersion that it actually reflects
gravitational waves (Press, 1 979).
For detailed analyses of absorption , scattering, and/or dispersion by
specific kinds of matter see the following references and the references cited
therein : black holes, Matzner et al. ( 1985); De Logi and Kovacs ( 1977);
neutron stars and other stars, Linet ( 1984); elementary particles, Section 7.2
of Zel'dovich and Novikov ( 1983); a magnetized plasma, Macedo and
Nelson ( 1983); a uniform medium, Esposito ( 197 la,b), Papadopoulos and
Esposito ( 1985), Szekeres ( 197 1), Section 4.2 of Grishchuk and Polnarev
( 1980), Section 2.4.3 of Thorne ( 1980); and a medium with boundaries,

Dyson ( 1969) , Carter and Quintana ( 1977).
Because gravitational and electromagnetic waves should propagate with
the same speed , they can interact in a coherent way (Gertsenshtein, 1962).
The interaction is so weak , however, that a substantial transformation of
one into the other requires propagation over a distance of order the radius of
curvature of the background spacetime which their own energy density
produces. Thus, such coherent interaction is not likely ever to be important
in the real universe - except possibly in gravity-wave detectors ; see Section
9.5.4(a) bel ow. For a review of the extensive literature on this subject see
Grishchuk and Polnarev ( 1980) . For a brief pedagogical discussion see
Section 17.9 of Zel 'dovich and Novikov ( 1983).
(b) Scattering by background curvature, and tails of waves

In regions where the background radius of curvature is comparable to or
shorter than the reduced wavelength , 9l8 ;5 X, the background strongly
scatters the waves. This is very important in some sou rces of waves, as the
waves are trying to form ; for example, it is responsible for the normal-mode
vibrations of black holes (Press, 197 1), and it leads to the formation of 'tails'
of the waves in a source's near zone (Price, 1972a ,b; Thorne, 1972 ;
Cunningham , Price and M oncrief, 1979) and to radiative tails in the wave
zone (Leaver, 1986b; Blanchet, 1987) - which, however, are not likely ever to
be observationally important.
(c) Gravitational focusing

Lumps of background curvature associated with black holes, stars, star
clusters, and galaxies will focus gravitational waves in precisely the same
manner as they focus electromagnetic waves ; and just as this focusing is
observationally important for the light and radio waves from a few very
distant quasars, so it might also be important for very distant discrete
sources of gravitational waves. Focusing by the sun, in the case of waves
with sufficiently short wavelength can be significant, but not at earth ; the
focal point lies farther out in the solar system, near the orbit of Jupiter
(Cyranski and Lubkin, 1974).
(d) Diffraction
Near the focal point of a gravitational lens the waves cease to propagate
along null rays and begin to diffract, thereby lessening the strength of the
focusing. The analysis of this is no different for gravitational waves than for
362 K. S. Thorne
electromagnetic or scalar waves, since polarization plays no important

role. Diffraction causes focusing to be significant only when the lens has a
gravitational radius 2 M that is larger than or of order the waves' reduced
wavelength ,t. For an order of magnitude discussion see, e.g. Section 2.6. 1 of
Thome ( 1983); for full details see Bontz and Haugan ( 198 1).
(e) Parametric amplification by background curvature

In regions of a dynamical spacetime (e.g. the expanding universe) in which
the characteristic wavelength ,t of gravitational waves is larger than or
comparable to the background radius of curvature �8, ,t � �8, the waves
can be parametrically amplified by interaction with the dynamical
background (Grishchuk, 1 974, 1975a,b, 1 977 ; Grishchuk and Polnarev,
1980). Viewed quantum mechanically, the interaction causes stimulated
emission of new gravitons. This effect may well have enabled the expansion
of the universe to amplify vacuum fluctuations from the big-bang singularity
(from the Planck time) into a strong, stochastic background of gravitational
waves today; see Section 9.4.3 (d) below.
(f) Non-linear coupling of the waves to themselves (frequency doubling, etc.)

Because general relativistic gravity is non-linear, there is a non-linear
coupling of gravitational waves to themselves ; and in principle this leads to
such non-linear conversion processes as frequency doubling. However, in
practice these effects are not important in regions where the waves are waves
(where � � 2). This is because, in such regions, the dimensionless amplitude
of the waves is very small compared to unity (equation (2 1 ) above). For a
more detailed discussion see Section 2 of Thorne ( 198 5).
(g) Generation of background curvature by the waves

The generation of background curvature by the stress-energy of the waves
(Isaacson, 1968b, MTW Section 35. 1 5) is important in cosmological models
in any epoch when the waves are sufficiently strong that their energy density
is comparable to that of matter; see, e.g. Hu ( 1978) and Chapter 17 of
Zel'dovich and Novikov ( 1983). It is also important in a 'geon' - i.e. a bundle
of gravitational waves that is held together by its own gravitational pull on
itself (Wheeler, 1962 ; pp . 409-38 of Wheeler, 1964 ; Brill and Hartle, 1964).
But geons surely do not exist in the real universe; they are only theoretical
entities, useful for exploring issues in fundamental physics.
Wave-produced background curvature is also important in the idealized
situation where one plane-fronted wave gets focused by passing through
another ('plane-wave collisi on') : the focusing itself is produced by the wave
generated background curvature. Moreover, if the waves are precisely
planar, a spacetime singularity forms at the focal plane (Khan and Penrose,
1 97 1 ; Szekeres, 1 972 ; Nutku and Halil , 1 977 ; Tipler, 1 980; Matzner and
Tipler, 1984 ; Chandrasekhar and Xanthopoulos, 1986; Yurtsever, 1987a),
and the generation of background curvature plays a key role in the
singularity. In the more realistic case (which, however, almost certainly does
not occur in the real universe except conceivably near the big-bang), in
which the waves are almost planar but die out slowly at large transverse
distances, if the transverse size is sufficiently large compared to the initial
wave amplitude, then the focusing probably still drives the amplitude up far
enough - before diffraction can act - to make background curvature
generation become strong and force a singularity to form (Yurtsever,
1987b).
9.3 .6 Wave propagation in an idealized, asymptotically flat universe

The split of wave propagation and wave generation into two separate
calculations is worrisome to those physicists who seek the highest levels of
rigor. Unfortunately, in analyses of waves in the real universe, where the
background spacetime is complex , the split is essential . Progress cannot be
made without it. However, the split can be avoided to some extent in an
idealized 'universe' , where the source of interest resides alone in an
otherwise empty and asymptotically flat spacetime.
In such an idealized universe, and only there, it has been possible to treat
wave generation and wave propagation simultaneously, with a single,
elegant formalism that spans the weak-gravity portions of the wave
generation region, and all of the l ocal wave zone and distant wave zone. In
my opinion, the nicest version of this unified formalism is a variant of the
post-Minkowski formalism due to Blanchet and Damour ( 1986) and
Blanchet ( 1987a). See Damour ( 1986) for a review of this and of earlier work
by others. This formalism is especially nice because, although it entails an
expansion in the strength of the gravitational field, the expansion has been
carried out to all orders, and a number of beautiful theorems have been
proved about it.
Also of great interest, elegance, and beauty, in an idealized asymptotically
flat universe, are expansions of the spacetime curvature along the outgoing
light cone in inverse powers of the distance to the source (Bondi, 1960 ;
B ondi, van der Burg and Metzner, 1962 ; Sachs, 1962, 1963 ; Penrose,
1963a,b; Newman and Penrose, 1965). Such expansions, carefully
364 K. S. Thorne
formulated and combined with conformal transformations that bring

'infinity' in to finite locations, reveal asymptotic structures of asymptotically
flat spacetime that illuminate the properties of gravitational radiation. For
recent reviews and references see Newman and Tod ( 1980), Walker ( 1983),
Schmidt ( 1979, 1986), Ashtekar ( 1984).
9.3 . 7 Exact, analytic solutions for wave generation and propagation
Those who seek high rigor (and even less rigorous people like me) have also
found pleasure in the few existing exact, analytic solutions of the Einstein
field equations that describe sources which radiate into asymptotically flat
spacetime (e.g . Bicak, 1968 ; Bicak, Hoenselaers and Schmidt, 198 3 ; Bicak,
1985; and references therein); and in idealized exact solutions for cylindrical
waves (Einstein and Rosen, 1936 ; Weber and Wheeler, 1957) and for planar
waves (e.g. Rosen, 193 7 ; Bondi, Pirani and Robinson, 1959; Ehlers and
Kundt, 1962 ; Sections 3 5.9-35. 12 of MTW). Also pleasing for its rigor is the
extreme limit of geometric optics, where the wavelength becomes so short
that the radiation is compacted into a gravitational shock wave (e.g. Pirani,
1957 ; Papapetrou , 1977 and reference� therein) .
9.4 Astrophysical sources of gravitational waves

In the real universe it is useful to divide the anticipated gravitational waves ,
into three classes, according to their temporal behaviors : bursts, which last
for only a few cycles, or at most for times short compared to a typical
observing run ; periodic waves, which are superpositions of sinusoids with
frequencies Ji that are more or less constant over times long compared to an
observing run ; and stochastic waves, which fluctuate stochastically and last
for a time long compared to an observing run.
In this section I shall describe the present knowledge and speculations
about various sources of gravitational waves. Attention will be restricted to
those sources which are most promising for detection by present or planned
detectors, with Section 9.4. 1 focusing on burst sources, 9.4.2 on periodic
sources, and 9.4.3 on stochastic sources.
As was emphasized in Section 9. 1 . 1 , for each source, with the single
exception of binary stars and their coalescences (Section 9.4. l(e) below),
either (i) the strength of the source's waves for a given distance from earth is
uncertain by several orders of magnitude; or (ii) the rate of occurrence of
that type of source, and thus the distance to the nearest one, is uncertain by
several orders of magnitude ; or (iii) the very existence of the source is
uncertain. Despite these uncertainties, it is important in the wave-detection
effort to estimate as best one can , for each source, the strengths of the waves
bathing the earth . Such estimates will be stated below, with references; and
they are collected together beiow In Figs. 9.6 (burst sources) , 9.7 (periodic
sources) and 9.6 (stochastic sources).
9 .4 . 1 Burst sources
(a) Bursts with and without memory

As Braginsky and Grishchuk ( 1 985) have recently emphasized,
gravitational-wave bursts can be subdivided into two classes : normal bursts,
in which h JkT begins zero before the burst and returns to zero afterward , and
bursts with memory, in which h ],! (by convention) begins zero before the
burst and then settles down into a non-zero, constant value .1.h r (the burst's
'memory') after the burst is over.
Physically, a burst with memory arises whenever the source, before the
burst or afterward or both, consists of several free bodies that are moving
with uniform velocities relative to each other. For example, the explosion of
a star into several pieces will produce a burst with memory, as will the
collision of two freely moving (not binary) stars or black holes, or the
gravitational scattering of a star by a black hole; but the birth of a black hole
in non-spherical stellar collapse will produce a burst without memory. In all
cases, the 'memory' . is the change in the total ' 1/r' coulomb-type
gravitational field of the source (Braginsky and Thome, 1987). For example,
in the quadrupole approximation , when one takes account of the fact that
one can add a time-independent constant to h ],! ('gauge change' that moves
stuff between the background field and the wave field) so as to make h ],! zero
initially, equations ( 12) and ( 1 3a) give
( ;
L\h ;.r = L\ � 4 A � V�
m rF (25)
Here the summation is over the free bodies in the system, m A and v� are the
mass and velocity of body A, and L\ denotes the change from before the burst
is emitted to afterward.
As is discussed by Braginsky and Thome ( 1987), the memory part of a
burst, L\hj,! , can be studied by any detector (with adequate sensitivity) that
operates at a frequency lower than the burst's characteristic frequency,
f ;:5 fc · Put equivalently, one can think of a burst with memory as having a
signal that extends down to all frequencies below fc (cf. the 'zero-frequency
limit' discussed by Smarr, 1977b and by Bontz and Price, 1979).
Current prejudice suggests that the strongest of bu rst sources (and thus
366 K. S. Thorne
the most interesting) may produce normal bursts rather than bursts with
memory ; but this prejudice could perfectly well be wrong. In accord with
this prejudice, the remainder of this section will focus on normal bursts.
(b) Characterization of the waves from a normal burst source and the noise in
a detector searching for them
Burst sources are best characterized by their full wave forms hjkT(t).
However, when comparing with detector sensitivities, it is helpful to have a
more compact characterization. Past discussions have used a loosely
defined 'characteristic amplitude' he and 'characteristic frequency' fe ·
However, the factors of order 3 that are glossed over by the loose definitions
are beginning to be important in the planning of gravity-wave searches,
especially in the case of the inspiral and coalescence of binary neutron stars
(subsection (e) below) ; and I therefore shall be quite careful in my definitions
of he and fe and subsequently in my corresponding discussions of detector
sensitivities.
As an aid to defining he and fe with care, I show in Fig. 9.2 a diagram of the
emission, propagation, and absorption of the wave. The source has
Fig . 9 .2. The angles I , /3, 1/1, e, <P which characterize the emission, propagation
and detection of a gravitational wave.
Sou rce
Po l a r i za t i o n
P l a n e· ---
x
preferred local Cartesian axes (x, y, z) with respect to which its internal
structure is especially simple. We shall denote by (z, /3) the direction toward
earth (spherical polar angles) relative to those axes. The detector, similarly,
has preferred local Cartesian axes (x, y, z) with respect to which its internal
structure is especially simple; and we shall denote by (8 , ¢) the direction
toward the source (spherical polar angles) relative to those detector axes.
The waves themselves, as they pass through the detector, are most simply
described in a third set of Cartesian axes (x', y', z') with origin at the
detector's center of mass , z' axis along the waves' propagation direction, and
x' and y' axes so oriented in the polarization plane as to make the wave
forms h + = h �!x, = - h �.;,. , and h = h ��· = h��· , especially simple. We shall
x
denote by t/I the angle between the x' polarization axis and the </> = 0 plane.
From a model for the source one can compute the wave forms
h + (t - z' ; z, /3) and h (t - z' ; z, /3) arriving at the detector. If the detector is
x
small compared to a reduced wavelength, as we shall assume throughout

Section 9.4 and in Sections 9.5. 1-9. 5.3 but not 9.5.4-9. 5.6, then it will feel a
simple linear combination of h + and h ; i.e. it will detect the wave form
x
h(t) = F + (8 , </>, t/l )h + (t; z, /3) + F (8, </> , t/l)h (t; z, /3) .
x x (26)
Here F + and F are detector beam-pattern functions (to be discussed in
x
Sections 9. 5.2 and 9.5.3 below), which depend on the direction of the source
(8 , <f> ) and the orientation t/I of the polarization axes relative to the detector's
orientation (Fig. 9.2) and which have values in the range 0 � I FA I � 1. There
will be some special choice of 8, </>, t/I for which F + = 1 and F = O ; we shall
x
call this the 'optimum source direction and polarization' for the + mode of
the waves.
In Section 9.5 we shall characterize the noise in a detector by a frequency
dependent spectral density Sh (f) (with dimensions Hz- 1 ) defined as follows :
if a precisely sinusoidal gravitational wave with known phase ex, known
frequency f > O and unknown rms amplitude h0,
h = (2)1h0 cos(2nft + ex), ex = con st. , (27a)
impinges on the detector, and if the experimenters seek to detect the wave by
Fourier analyzing the detector output with a bandwidth !J.f (integration
time f = 1/flf), then the amplitude signal-to-noise ratio will be
S h0
- (27b)
N [Sh (f)�J]f
In much of the gravitational-wave literature this Sh (f) is denoted [h(f)] 2 ;
i.e. h(f) = [Sh (f)] 1 (with dimensions Hz-!).
368 K . S. Thorne
In this section we are interested not in periodic gravitational waves, but

rather in bursts - i .e. in waves with complicated time dependences h(t) and
with durations short compared to the experimenters' observation time. We
shall assume that the wave form h(t) is known, and that our only question is
whether the wave i s really present or not - with some postulated starting
time t0• (Later we shall discuss the statistical consequences of an unknown
starting time.) There is a well-known optimal strategy for searching for such
a wave in the presence of noise with known spectral density :
(i) One constructs a Wiener optimal filter K(t), that function of time whose
Fourier transform K (f) is the same as the transform n(f) of the signal h(t),
oo
weighted by l/Sh(f) (so that noisy frequencies are suppressed) :
h
K (f) = (f) ' f-+oo h(t)
h(f) = e i 2 /r d t ,
tt (28a)
oo
Sh(f)
K(t) = f- K (f) e -i2ttfr
+oo df.
(ii) One then takes the output of the detector, which includes noise and
(28b)
possibly signal ; from it one computes the gravitational wave form h output(t)
that would have been required to produce the observed output if there
oo-+ K(t - t0)houtput(t) dt.

were no noise present ; and one then computes a quantity
W=
f oo
(iii) This quantity will have root-mean-square contribution N from noise ;
( 2 8c)
and if the signal was actually present with starting time t0 , it will have a
signal contribution S (i.e. W = N + S), with squared signal-to-noise ratio
s2
N 1=
J.00 2ji1(f)l
o Sh(f)
2
df. (2 9)
See Wiener ( 1949), Sections 25-27 of Wainstein and Zubakov ( 1 962) , Kafka
( 1 977), and Michelson and Taber ( 1 984) for proofs and discussion. The
integral in (29) is from 0 to oo rather than - oo to + oo and there is a factor 2
present because Sh(/) is a 'one-sided spectral density' (defined only for f";::. 0 ;
negative frequencies are folded into positive). The squared signal-to-noise
ratio (29) will be the basis for our definitions of he and fc ·
Expression (29) is the squared signal-to-noise ratio for a specific
('fiducial') source at some specific distance r0 from earth . Suppose that space
is uniformly filled with sources, all identical to this fiducial source but with
random directions ( 8 , ¢), orientations (z, /3), and polarization angles t/J.
Suppose, further, that inside the source's distance r0 there is, on average, one
bu rst each D0 days. Then what, on average, will be the squared signal-to
noise ratio (52/N2)51rongest of the strongest burst that occurs each D0 days?
One might expect the answer to be the fiducial 52 /N2 of (29), averaged over
all the angles e, </J, I/I, z , /3. Not so. There is a statistical preference for
directions and polarizations that give larger values of 52 /N2, because they
can be seen out to greater distances where the event rate is greater. This
effect gives, assuming the event rate goes up as r 3 and h ],! goes down as 1 /r ,
(52/N2)strongest = (53 /N 3)i where ( · ) denotes an average over randomly
" "
distributed angles. Now, the 1 power is a pain to deal with in subsequent

calculations, so we shall switch to a straightforward angular average
(52/N2 ) and to compensate we shall insert a multiplicative factor �, which is
(approximately) the ratio of (53/N3)i to (52/N2) if both source and
detector have quadrupole beam patterns ('slow-motion, poor-antenna
regime'). This, together with (F� ) = (F: ) and (F + F x ) = O (true for any
quadrupole-beam-pattern gravity-wave detector) , permits us to write
(52 /N2)strongest in the following approximate form, where we omit the
subscript 'strongest' for ease of notation and where <lii+ l 2 + liix l 2 ) is
averaged over source angles (1, /3) :
(30)
We shall now specialize , for the remainder of this subsection, to gravity

wave searches with broad-band detectors, i.e. detectors for which the noise
5h (f ) is small over a band of frequencies !min � f � !max ' with fmax � 2fm in ·
Narrow-band detectors will be treated separately in Section 9.5 .2(b) below.
For broad-band detectors equation (30) motivates us to define the
characteristic frequency and amplitude of our fiducial source, at distance r0,
by
(3 1a)
(3 1b)
where the average is over randomly distributed source orientation angles z ,

f3. Similarly, we shall define a characteristic detector noise amplitude at
frequency f by
(32)
370 K . S. Thorne
In terms of these quantities equation (30) reads

S he
(33)
N h n (fe) '
This is the S/N of the strongest burst that arrives at earth, on average, at the
same rate as bursts occur inside the fiducial source 's distance r0•
In Fig. 9 .4 below we characterize detector burst sensitivities not by h n (f ),
but rather by a quantity h 3/ yr that answers the following question : What is
the characteristic strength he = h 3 / y r of a source with sufficiently large S/N
(equation (3 3)) that, if it is seen three times per year (once each 10 7 seconds) by
two identical detectors operating in coincidence, we can be 90 % confident the
detectors are notjust seeing their own noise. The use of two detectors permits
(we shall presume) the elimination of non-Gaussian noise. Then the
Gaussian distribution of the amplitude noise, together with the use of two
detectors and the fact that in i = 10 7 s the experimenters must try roughly
2nfci :::::. fe / 10 - 8 Hz starting times t0 for their Wiener optimal filter, implies
that we require S/N :::::. [ln(fe / 10 - 8 Hz)] 1 ; and, correspondingly,
,
[(
h 3 1 , "" i n
10 !, Hz ' )J (}{(::�:)��) l "" (3 -S )h . (fc). (34)
Here the range 3-5 corresponds to the range of frequencies of interest for
most burst searches, 10 - 4 Hz- 10 + 4 Hz. Note that because the events being
sought are so far out on the tail of the Gaussian probability distribution,
changing by a factor 10 or 100 the number of trial starting times t0, or asking
for 99 % or 99.9 % confidence rather than 90 %, would have a negligible
effect on this h 3 / y r ·
It is often useful to rewrite the fc and h e of equations (3 1) in terms of the
energy flux per unit frequency dEGw /dA df carried past the detector by the
waves. From equation ( 10) for the waves' stress-energy tensor together with
Parseval 's theorem we infer that
dEGw n
dA df 2 2 ( 1 - + ( ) 1 2 1 - x ( ) 1 2
= f h f + h f ), (35)
where the extra factor 2 comes from folding negative frequencies into
positive (so f > 0). When this quantity is averaged over all directions z, {3 , it
gives (4nr;) - 1 dEGw/df, where r0 is the distance to the source ; and
consequently equations (3 1) become
· [f co
-t
J [f
co
J
dEGw/df dEGw /df
fe = d ln f d ln f , (36a)
_ co fSh (f) _ co Sh (f)
(36b)
For most burst sources (e.g. supernovae) the wave form is so uncertain
that a careful calculation of he is unjustified. In such cases it is useful to
reexp ress he (equation (36b )), approximately , in terms of the total energy
( lO )
/j.Eaw radiated :
/j.£a�/ )
(� /j. ! !
he '.:!:. !c 1 = 2 . 7 x 10 _(
20 a� E ) ( l kHz ) Mpc
2
n r0 M 0c fc r0
, (37)
where M 0 is the mass of the sun and 10 Mpc is the distance to the center of
the Virgo cluster of galaxies (assuming a Hubble constant of 100 km s - 1
Mpc - 1 ).
We turn now to a discussion of specific burst sources, and their
characteristic frequencies fe and wave strengths he.
(c) Supernovae (collapse to neutron star)
Supernovae of 'type II' are believed, with a high level of confidence, to be
created by the gravitational collapse, to a neutron-star state, of the cores of
massive, highly evolved stars. Supernovae of 'type I', by contrast, are
thought to result from nuclear explosions of white dwarfs that are accreting
mass from close companions - explosions in which the stellar core probably
does not, but might, collapse to a neutron-star state (Woosley and Weaver,
1986; Evans, Iben and Smarr, 1987, and references therein) . In addition to
these two types of optically observed supernovae, there may well be stellar
collapses to a neutron star that produce little optical display ('optically
silent supernovae'). _
The rate of occurrence of supernovae of types I and II is fairly well

determined observationally : in our Galaxy roughly one type I each 40 years
and one type II each 40 years ; out to the distance of the center of the Virgo
Cluster of Galaxies ( 10 Mpc) several type I and several type II per year; rate
increasing roughly as (distance)3 near and beyond Virgo; see, e.g. Tammann
( 198 1). Thus, to have an interesting event rate one must have adequate
sensitivity to reach the center of Virgo, about 10 Mpc. Optically silent
supernovae could be more frequent, since statistics on normal stars massive
enough to form neutron stars when they die, permit a neutron-star birth rate
that could be as high as one every four years in our galaxy if mass loss in the
late stages of stellar evolution is smaller than normally thought.
Alternatively, it is conceivable that optically silent supernovae never occur.
The strengths of the waves from a supernova depend crucially on the
372 K . S. Thorne
degree of non-sphericity in the stellar collapse that triggers it, and somewhat
on the speed of collapse - i.e. on whether the collapse is nearly free-fall ('cold
collapse') or is more gentle due to the resistance of thermal .pressure ('hot
collapse'). Perfectly spherical collapse will produce no waves; highly non
spherical collapse will produce strong waves. Little is known about the
degree of non-sphericity in type II (which are surely due to stellar collapse) ,
but current prejudice suggests that the typical type II might be quite
spherical and thus poorly radiating. If type I are due to explosion of an
accreting white dwarf and, contrary to current thought , the explosion is
accompanied by collapse of the stellar core to a neutron star, then the white
dwarf might be rapidly rotating due to the accretion, and centrifugal forces
might then cause it to collapse very non-spherically and radiate strongly.
In the mid- 1970s there was a swing of fashion from believing that the
collapse is cold and fast to believing it is hot and slow (e.g. Wilson , 1974 ;
Schramm and Arnett, 1975). Some astrophysicists were aghast at the
consequence of this swing : for example, in the highly non-spherical but
axisymmetric collapse models of Saenz and Shapiro ( 1978 , their tables 1-4)
the total energy radiated as gravitational waves was reduced from
l!! EGw /M 0c2 "" 6 x 10 - 3 for cold and fast to "" 1 x 10 - 5 for hot and slow.
This boded ill . for attempts to detect gravitational waves, the astrophysicists
thought. However, closer scrutiny revealed ·relatively little change in the
prospects for detection : the total energy radiated dE Gw is a rather poor
indicator of detectability. Much more relevant is the amplitude signal-to
noise ratio S/N = h c /hn(fc) (equation (33)). The planned LIGO beam
detectors, when optimized for frequency fc , have hn(fc) oc fc (Fig . 9.4 and
equation ( 12 5a)) ; and this together with equation (37) gives
S/N oc (dE Gw /fc3)1 . (38)
Although the new fashion (hot and slow) corresponded to a reduction in
dEGw by a factor 600, the characteristic frequency of the waves also went
down (from � 3000 Hz to � soo Hz; see Saenz and Shapiro , 1978) ; and,
correspondingly, S/ N was reduced by less than a factor 2 .
This illustrates the importance of thinking in terms of h e , fc , and S/N =
hc /hn(fc ) rather than in terms of energy radiated, when evaluating the

strengths of gravitational waves.
Corresponding to our poor knowledge of the strengths of the waves from
supernovae is a similar poor knowledge of the wave forms. It seems unlikely
that theorists will be able to firm up their predictions of the waveforms
before observers detect and study them. Thu s, it is best to think of the
computed wave forms as giving us a tool for translating future observations
into an understanding of what is happening in the stellar core. As an

example consider Fig . 9.3, which shows the wave forms computed· by Saenz
and Shapiro ( 1978 , 1 98 1), using the quadrupole formalism, for two different
stellar collapses. In each case - and in general, when one thinks the
quadrupole formalism may be accurate - one can invert equation ( 12) to see
how the source's quadrupole moment is behaving. The left curve (Saenz and
Shapiro , 1978) shows several epochs labeled FF in which h + (t) varies
approximately as I t - t0l - i , corresponding to free-fall motion ; and these free
fall epochs are separated by three brief periods with sharply reversed peaks
(labeled 'P' in the diagram) corresponding to a sharp acceleration in the
direction opposite to the free fall . Considering the timescales of hundredths
of a second for the free-fall epoch (characteristic of late stages of collapse to a
neutron star) and "' 0.5 milliseconds for the peak epochs (characteristic of
neutron-star pulsation), the natural and correct interpretation is that these
waves are from collapse to a neutron star in which the stellar core bounced
sharply three times. The fact that the three sharp peaks are all in the same
direction (up , not down) indicates that the sharp bounces were all along the
same axis. Surely the other axis that projects on our sky should have
bounced as well , or at least stopped its collapse; so there should be at least
one sharp peak in the down direction. Indeed there is ; it is superposed on the
central up peak (region labeled E in the diagram). The natural and correct
interpretation is that the star was centrifugally flattened by rotation ; its pole
collapsed fast and bounced three times (up peaks P) while its equator
collapsed more slowly and bounced once (down peak E).
Fig. 9 . 3 . Wave forms produced by two very different scenarios for the
collapse ofa normal star to form a neutron star . Wave form (a) is from Saenz
and Shapiro ( 1978); (b) is from Saenz and Shapiro ( 198 1).
1.2 1.0 ��-�-..,..-�-...----...,....---.--,
1.0 .8
.8 p .6
.6 • .4
:4 .2
h .2 h 0 1------...
0 -.2
-.2 .,('-
.,('- - .4
-:4 - .6
-.6 -.8
-:8
- 1 .0 L...-'--_.__---'--.l..-.L..----'--�---'..l.
4 6 0.2 0 .4 0.6 0.8
2
-2 2
Tim e , 10 sec Tim e , I 0- se c
(b)
(a)
3 74 K . S. Thorne
The very different right-hand curve in Fig. 9.3 (Saenz and Shapiro, 198 1)
implies a quadrupole moment that somehow is driven into sinusoidal
oscillations which initially increase in amplitude and then die out. Again, the
fact that the period ( - 0.6 ms) is that of a neutron-star pulsation suggests
that something is triggering, then damping, such a pulsation. If this wave
form was seen roughly one day before an optical supernova was found in the
Virgo cluster, one would infer that the waves were from the supernova and
one could deduce that the quadrupole-moment oscillations are so large in
amplitude that they must have absorbed, say, 10 - 4 of the collapse energy.
The natural explanation might be parametric amplification of quadrupole
neutron star pulsations by a bouncing stellar collapse, followed by
hydrodynamic damping - the process that gave rise to this computed wave
form.
A large amount of effort has gone into model calculations of gravitational
collapse to a neutron star and the waves it emits ; but the effort has not
produced a consensus by any means! For a detailed review of the literature
up to 1982 see Eardley ( 1983) ; for an update on that review see M iiller
( 1984). In the case of rapidly rotating collapse, where the emission should be
strongest but the event rate is totally unknown (most collapses could be
slowly rotating), there are three radically different scenarios and
corresponding wave characteristics : (i) the star may remain axisymmetric
throughout the collapse. In this case the best current 'wisdom' (but by no
means a consensus) comes from calculations by Miiller ( 1982) and is
pessimistic. Those calculations predict the strongest emission to come in two
different spectral regions : fc - 1000 Hz where l!iE Gw "' 1 x 10 - 7 M 0 c 2 and
hc - 1 x 10 - 2 3 ( 10 Mpc/r0) due to the initial collapse and bounce ; and
fc - 104 Hz where �EGw "' 10 - 6 M 0c 2 and hc - 1 x 10 - 2 3 ( 10 Mpc/r0) due to
pulsations of the newly formed neutron star.- (ii) The star may become
unstable to an 'm = 2 bar-mode' deformation so it rotates end-over-end like
an American football. In this case the best current 'wisdom' is more
optimistic : the calculations of Ipser and Managan ( 1984) predict a highly
monochromatic emission at f - 1000 Hz, lasting for - 30 cycles and
producing l!iEGw "' 3 x 10 - 4 M 0 c 2 and hc - 5 x 10 - 22 ( 10 Mpc/r0). (iii) The
collapsing star may become so strongly unstable to non-axisymmetric
perturbations that on the way down it breaks up into two or more discrete
lumps. Very little is known about this possibility, and radiation reaction
from the m = 2 mode (case ii) might prevent it from occurring at all (Ipser,
1986). Eardley ( 1983) argues that if it does occur, it may produce quite
strong waves : l!iEGw - (a few) x 10 - 2 M 0 c 2 at fc - 1000 Hz, corresponding
to hc - 4 x 10 - 21 ( 10 Mpc/r0) .
Because this best wisdom is so insecure, Fig . 9.4 shows wave strengths
based not on these specific models, but rather on the general equation (37)
for several possible values of fl.EGw and r0, and for the entire range of
characteristic frequencies that have shown up in model calculations,
200 Hz :S fc :S 10 000 Hz. Note that detectability depends strongly on
whether the waves come off at low frequencies or high : a factor 8 reduction
in fl.EG w can be compensated by a factor 2 reduction in fc ·
Fig . 9 .4. The characteristic amplitudes he (equation (3 lb)) and frequencies fc

(equation (3 la)) of gravitational waves from several postulated burst sources
(thin curves) , and the sensitivities h3;yr of several existing and planned
detectors (thick curves and circles) (h3;yr is the amplitude he of the weakest
source that can be detected three times per year with 90 % confidence by two
identical detectors operating in coincidence). The abbreviations BH, NS and
SN are used for black hole, neutron star and supernova . The sources are
discussed in detail in the indicated subsections of Section 9.4. 1 , and the
detectors in the indicated subsections of Section 9.5.
Doppler
)-
-5.5' o
G PS
\
\
10 Me Hole
376 K. S. Thorne
We note in passing that neutron-star pulsations might be excited not only

as part of the star's birth throes, but also as a consequence of a sudden strain
release or phase transition in an old neutron star (Thorne, 1 9.78 ; Ramaty et
al., 1980; Haensel , Zdunik and Schaeffer, 1986). Since there are 10 8-109 old
neutron stars in our galaxy, it is conceivable - though not highly likely - that
our galaxy could produce an interesting event rate. To be detectable by the
planned LIGOs, the waves would need to have he � 10 - 2 1 at fc "" 3000 Hz
corresponding to
(39)
(d) Collapse of a star or star cluster to form a black hole
As with collapse to form a neutron star, so also for collapse to a black hole,
the strengths of the waves produced are highly sensitive to the degree of non
sphericity, and the typical degree of non-sphericity is unknown. Equally
unknown in the black-hole case, by contrast with the neutron star, is the
frequency of occurrence of such collapses :
It is very likely that black holes exist in our universe with masses
throughout the range 2 M 0 � M � 10 1 0 M 0 (see Chapter 8 of this book).
The holes of lowest mass can only form by direct collapse of a star. Those of
higher mass, however, can form by many routes (direct collapse; gradual
growth from a small hole by accretion ; collision and coalescence of smaller
holes ; . . . ). For discussions of the routes that might occur in a dense galactic
nucleus see Blandford ( 1979) and Rees ( 1983). Which routes actually occur
and how often are almost totally unknown. However, roughly known upper
limits on the birth rates of the smallest and the largest holes give - under the
assumption that all births are by direct collapse - corresponding upper
limits on the rate at which gravity-wave bursts from such collapses hit the
earth. For holes of a few solar masses the birth rate under reasonable
assumptions (Section 8 .3 . 3 of Chapter 8) should not exceed ,...,, 1 the birth
rate of neutron stars; and correspondingly, at the distance of the Virgo
cluster it should not exceed '""" 1/year. Bethe ( 1986) argues that the rate may
actually be of this magnitude. At the other end of the spectrum , holes with
masses M � 106 M 0 probably occur only in galactic nuclei ; and over its
lifetime each galactic nucleus might give birth, at maximum , to only a few
such holes. Correspondingly (Thorne and Braginsky, 1976 ; Blandford ,
1979) , the maximum rate of collapse-births of supermassive holes,
M ;(; 106 M 0 , is a few per year throughout the observable universe - i .e. out
to the Hubble distance. It is fashionable to believe that the actual rate is
much less than this upper limit (e.g. Rees, 1983).
In one respect collapse to a black hole is better understood than collapse

to a neutron star : the final object is far simpler, and correspondingly the
waves from its vibrations, if they are triggered by the collapse, are far better
understood. Detailed calculations suggest, in fact, that black-hole
vibrations are rather easy to trigger (e.g. Detweiler, 1977) and that when they
are triggered , the most slowly damped one or two quadrupole modes will
dominate. Thus, while the details of the initial burst of waves may depend on
unknown details of the collapse, the late-time behavior will have a well
established damped oscillatory form from which one can read off the mass of
the hole with excellent accuracy and its angular momentum with modest
accuracy (Detweiler, 1980; Leaver, 1985, 1986a ; Stark and Piran , 1986;
Piran and Stark, 1986).
As a specific example, Fig . 9.5 shows the waves produced by a specific
model of a collapsing, rotating star as computed using numerical relativity
techniques by Stark and Piran ( 1986). The solid curve is the computed
waveform, and the dashed curve is a fit to it using a mixture of the two most
weakly dampled quadrupole modes of a non-rotating hole. The fact that the
fit is so good shows (i) that the final ringdown is, indeed , due to the black
hole vibrations ; and (ii) that the hole was not rotating extremely rapidly, i.e.
it had ( 1 - a/M) <; 0.3, where a is the specific angular momentum and M the
mass. If the hole had had 1 - a/M < 0.3, the 'Q' of the hole's oscillations
would have been noticeably larger, and the ringdown ,of the waves would
have been noticeably slower. (For details of the normal modes of black holes
and the frequencies and damping times of the waves they should produce see
Detweiler, 1980, and Leaver, 1985, 1986a.)
If collapse to a black hole radiates with an efficiency dE/Mc 2 = e and the
hole is at a distance r0 and has a mass M, then the characteristic frequency
and amplitude of its waves will be (Stark and Piran, 1986; Piran and Stark,
i(':o )
1986; and equation (37) above)
I
J, = ( 1 .3 x 10• Hz . (40a)
( ) ( ) ( )( )
5 n: M
-
15 t M e ! M 10 Mpc
h c � - e - = 7 x 10 _ 22 0.0 1 M0
( ) ( )( )
r0
- -
- 2n r0
_ 2 0 e ! 10 3 Hz lO Mpc
_
- 1 .0 x 10 • (40b)
_
0.01 fc r0
If the collapse is axisymmetric, then the efficiency e probably does not exceed
7 x 10 - 4 (Stark and Piran, 1986). However, in the non-axisymmetric case
(e.g. formation of an elongated configuration due to rapid rotation, or
378 K. S. Thorne
bifurcation into one or more lumps during collapse (the 'collapse, pursuit,
and plunge' scenario of Ruffini and Wheeler, 197 1)), the efficiency might be
in the range 0.0 1----0 . 1 (see, e.g., Eardley, 1983, and Rees, 1983). The source
characteristics (40) are shown in Fig. 9.4 for black-hole births at the Hubble
distance and at the distance of Virgo, with efficiencies of c = 10 - 2 and 10 - 4•
(e) Coalescence of compact binaries (neutron stars and black holes)

Since a large fraction of all stars are in close binary systems, the dead
remnants of stellar evolution may contain a significant number of binary
Fig. 9.5. The gravitational wave fo rm produced by the gravitational collapse

of an axisymmetric rotating star to produce a Kerr black hole, as computed
by Piran and Stark ( 1986) using numerical relativity techniques. The star
and the black hole it forms both have J/M 2 =. a/M = 0.63 (where J is angular
momentum and M is mass). The dashed curve is a fit, to the wave fo rm, of a
superposition of the waves from the two most slowly damped quadrupolar
+
iwi r
normal modes of a non-rotating hole with mass M, h + "" Real { A 1 e -
A 2 e - iw2 1 } with
w1 = (0.374 - 0 .089i)/M , w 2 (0.348 -0 .274i)/M . The fitting
-0.9 - 1 . li, A 2 0.9 + l .4i. Thus, these two modes are
=
amplitudes are A1 = =
roughly equally excited b y t h e collapse.
- - - - mode f it
-- ca l cu l a t ed
...)
(\J
c
+
·-
.....c.
-. 0 4 .___.__........
._ ..._
._ ..._
_ ....i.
_ _,__....
...._ -L...
.L... ....
.. ....
.J.... ....
.J.... --L--L..--L..
.J.... _J
-20 0 20 40
retarded time ( t-z')

systems whose components are neutron stars or black holes, and are close
enough together to be driven into coalescence by gravitational radiation
reaction in a time less than the age of the universe. The binary pulsar
PSR 1 9 1 3 + 16 is an example of such a system ; it will coalesce 3 . 5 x 108 years
from now.
As the two bodies in a compact binary spiral together, they emit periodic
gravitational waves with a frequency that sweeps upward toward a
maximum,
fma x � 1 kHz for neutron stars, (4 1a)
10 kHz .
for holes with the larger having mass M 1 . (4 lb)
M 1 /M
fmax �
0
The wave form during the nearly Newtonian part of the frequency sweep ,
f � fmax • is easily computed from the quadrupole formalism (Section 9.3.2).
The post-Newtonian corrections to this waveform will become more and
more important as f rises toward fmax ; they are given in Wagoner and Will
( 1976) ; see also Gal'tsov, Matiukhin and Petukhov ( 1980). Ultimately, near
fm a x ' higher-order corrections or full non-linear relativity are needed to get
the wave form reasonably accurately. The final, coalescence stage will be
especially interesting and complex in the case of a neutron-star binary, and
may be quite sensitive to the masses of the two stars; and as with
supernovae, we might not understand reliably what to expect until
gravitational-wave observations show us. For a fi rst , preliminary
theoretical effort at understanding, see Clark and Eardley ( 1977). For black
holes, by contrast, numerical relativity is likely to give us, within the next
five years or so , a detailed and highly reliable picture of the final coalescence
and the wave forms it produces, including the dependence on the hole's
masses and angular momenta . Comparison of the predicted wave forms
with observed ones will constitute the strongest test ever of general
relativity. (The wave forms for the astrophysically unlikely cases of head-on
collisions of two identical non-rotating holes or neutron stars have already
been evaluated by numerical relativity ; see Smarr, 1977a for black holes,
and Gilden and Shapiro, 1984 for neutron stars.)
Because the binary system spends far more time in the early, low
frequency part of the sweep than in the later, high-frequency part or in the
final coalescence, and because planned gravity wave detectors have less
amplitude noise at low frequencies, f- 100 Hz, than at high, f � 100 Hz (cf.
Fig. 9.4), it will be easier for detectors to see the Newtonian regime of the
sweep than the post-Newtonian regime or the final coalescence - except in
380 K. S. Thorne
the case of black-hole binaries with M 1 ,...., 100-1000 M 0 or M 1 � 10 6 M 0

(Fig. 9.4). In the Newtonian regime, if we orient the polarization axes ex' and
ey' along the major and minor axes of the projection of the orbital plane on
the sky, then the wave form will be
h + 2( 1 + cos2 i)(µ/r)(nMJ)13 cos(2nft),
= (42a)
h " = + 4 cos i(µ/r)(nMJ)1 sin(2nft). (42b)
Here it is assumed that the orbit is circular because radiation reaction long
ago will have forced circularization (Peters and Mathews, 1963) ; l is the
angle of inclination of the orbit to the line of sight ; M and µ are the total and
reduced masses
(42c)
and f, the frequency of the waves (equal to twice the orbital frequency) , is
[ iJ
given as a function of time by (MTW equation (36. 17))
1 5 1 1
f= . (42d)
n 256 µMi (t0 - t)
The most promising detectors for coalescing neutron-star binaries and
low-mass black-hole binaries are beam detectors in the planned multi
kilometer LIGOs. As we shall see in Section 9.5.3(e), a beam detector can be
operated in several different optical configurations. The optimum
configuration for searching for coalescing binaries is likely to be one with
light recycling, for which the spectral density of shot noise (the dominant
noise above some 'seismic cutoff frequency fs) will have the form
Sh(f) = const x fk[ l + (f/fk) 2 ] at f > fs· (43a)
Here fk is a 'knee frequency' which the experimenters can adjust by
changing the reflectivities of certain �irrors in their detectors; see equation
( 1 17c) and Fig. 9. 13 below, and associated discussion. The constant in
equation (43a) is independent of the choice of fk· At frequencies below the
'seismic cutoff' fs seismic noise is likely to come on very strong ; accordingly,
we shall make the approximation
(43b)
By Fourier-transforming the wave forms (42), squaring, and averaging
()
over the source orientation angle i, we obtain
n µ 2 M3 1
< I h- + 1 2 + 1 h- " 1 2 ) (44)
12 -;: µ (nMJ)1 '
=
By inserting this source strength (44) and the detector noise (43) into
equation (30) and maximizing the resulting signal-to-noise ratio with
respect to the knee frequency fb we find that the experimenter will do best to
choose
(45)
For smaller choices of fk there is not a wide enough frequency band between
the seismic cutoff and the knee to take optimal advantage of the broad-band
nature of the signal . For larger values of fk the experimenter loses because
the height Sh (f) of the 'noise floor' at fs < f� fk (equation (43a)) is
proportional to fk· With this choice of knee, equations (3 la) and (43)-(45)
give for the characteristic frequency fc = 0.909fk · Below (equation ( 125a)) we
shall characterize the sensitivites of beam detectors in full scale LIGOs,
when searching for bursts, by the noise amplitude h n at the knee; and ,
correspondingly, we here shall set fc (which after all is somewhat arbitrary)
to fk rather than 0.909fk :
fc = Jk = 1 .44fs · (46a)
Equations (3 lb) and ( 43)-(45) then give for the characteristic amplitude of
the waves from inspiraling binaries
hc = 0.237 �
µ1 ! ( )( )(
= 4. 1 x 10 - 22 � )(
! M j lOO Mpc lOO Hz i ).
r0 fc M0 M0 r fc I
(46b)
=
The characteristic amplitude (46b) is enhanced over the actual rms wave
strength (h� (tc) + h : (tc)) 1 the waves have at the time t tc when they sweep
-
through frequency fc enhanced by very nearly the square root of the
number of periods, n = (f 2 /j)1 = fc = (5/96n)(M/µ)(nMfc) - 1 , that the
binary spends in the vicinity of the frequency fc· This Jn -:::::.
28(µ/M 0 ) - 1 (M/M 0 ) - ! (fc/ 100 Hz) - � enhancement corresponds to the
enhancement in effective signal that the experimenters will achieve by
optimal signal processing in their search for these frequency-sweeping
bursts.
From a study of the waveform (42) using broad-band detectors at several
widely spaced locations on the earth, one can deduce the following
information : (i) the direction to the source (which comes from phase
differences in the signal at different detectors in different locations); (ii) the
inclination of the orbit to the line of sight (which comes from the amplitudes
in the two different polarization modes) ; (iii) the direction the stars move in
their orbit (which comes from the + or - sign in equation (42b )); (iv) the
combination (µ 3 M 2 )! of the reduced and total masses; and (v) the distance r
to the source. If the mass combination (µ 3 M 2 )� is � l .5M 0 , one can be
382 K. S. Thorne
fairly sure the binary was made of neutron stars ; if it is much larger, one can
be fairly sure that at least one of the bodies was a massive black hole.
Especially intriguing is the possibility (Schutz, 1986b) that, .i n the case of
neutron stars the coalescence will produce electromagnetic emission (e.g .
due to an explosion of the less massive star, Blinnikov et al. , 1984) that is
strong enough to be detected at earth and thereby to pin down the source's
location with far higher precision than can be obtained from the
gravitational waves. In this case, a redshift will probably be obtainable from
optical observations; and that redshift together with the gravitational-wave
determined distance r will give a value for the Hubble constant. Schutz
(1986b), from a detailed study of the expected noise in future beam detectors,
concludes that the prospects are good thereby to obtain a significantly better
value for the Hubble constant than we now have. Even in the absence of
electromagnetic signals from the coalescence, it may prove possible by
statistical means to determine the Hubble constant using combined data for
a number of coalescences ; see Schutz (1986b) for details.
Clark, van den Heuvel and Sutantyo ( 1979) have estimated, from
neutron-star observations in our own galaxy, that to see three coalescences
of neutron-star binaries per year one must look out to a distance of
100 � i8° Mpc, where the quoted uncertainties are at the 90 % confidence
level . Correspondingly, in Fig . 9.4 are shown the characteristic amplitude
and frequency, for a range of values of the seismic cutoff fs and
corresponding values of fc l .44fs, produced by the coalescence of two 1 .4
=
solar mass neutron stars at 100 Mpc (estimated event rate about three per
year) and at 1 the Hubble distance (event rate about ten per day). Because
much has been learned observationally about the statistics of neutron-star
binaries since these estimates of event rates were made, a careful restudy of
the estimates is much needed.
As Fig. 9.4 shows, future earth based beam detectors may be able to see
black-hole coalescences throughout the universe, so long as the more
massive of the two holes does not exceed 1000 M 0/( 1 + z) , where z is the
hole's cosmological redshift. The coalescence rate for black-hole binaries of
a few solar masses could be of order that for neutron-star binaries (a few per
year at 100 Mpc), or might well be far lower. Particularly intriguing is a
scenario, suggested as very plausible by Shapiro and Teukolsky ( 1985) and
Quinlan and Shapiro ( 1 987), in which a large fraction of galactic nuclei
create, at some phase of their evolution , a dense cluster of neutron stars and
small-mass black holes which - on a timescale of only a few years - form
tight binaries that coalesce, with the coalesced holes then forming new tight
binaries that coalesce, . . . until the cluster goes unstable and collapses to
form a single large hole. This scenario suggests that in typical years the earth
might be hit by a number of spiraling wave bursts from coalescing binaries of
masses 3 M 0 -1000 M 0 at the Hubble distance, z ,.., 1 . Also intriguing but
much less likely is a scenario discussed by Bond and Carr ( 1984) in which a
sizable fraction of the mass of the universe is in black-hole binaries with
masses "" 100-1000 M 0 for which the coalescence rate could be several per
year in the local group of galaxies (distance - 1 Mpc).
One or more black hole binaries of any mass up to "" 108 M 0 might have
formed in the nuclei of a reasonable fraction of all galaxies during the past
-
life of the universe, leading to event rates <: I/year out to the Hubble
distance (cf. Fig . 1 of Rees, 1983) or they might never form. There is
actually observational evidence for supermassive black-hole binaries
formed by the coalescence of galactic nuclei (Begelman, Blandford and Rees,
1980 ; Rees, 1983) ; but the rate of such events probably does not exceed 1 60
years out to the Hubble distance (Rees, 1983).
(f) The fall of stars and small holes into supermassive holes
The supermassive (M 1 <: 105 M 0) black holes thought to inhabit the nuclei
of galaxies might typically grow by accretion on timescales as short as 108
years ; see, e.g . Section 8 .6 of Blandford and Thorne ( 1979). When such a
hole grows larger than 109 M 0 , normal stars can pass near or plunge
through its horizon without being torn apart tidally, and the number of stars
that so scatter or plunge could well be of order one per year or more (e.g.
Dymnikova , Popov and Zentsova, 1982) . For smaller supermassive holes,
scattering or plunging normal stars will be tidally disrupted, reducing the
strength of their waves; but the reduction will not be great, at least in the
case of radial infall , unless the hole is below 10 6 M 0 (Nakamura and Sasaki,
198 1 ; Haugan, Shapiro and Wasserman, 1982). For any hole, neutron stars
and satellite holes can scatter or plunge through without enough disruption
to strongly supp ress their radiation ; but the event rate (per supermassive
hole) will typically be well below one per year.
The wave forms emitted when a star or small hole is scattered by or
plunges into a supermassive hole have been evaluated with high precision
using perturbation formalisms; see, e.g . Detweiler and Szedenits ( 1979),
Kojima and Nakamura ( 1984a). The characteristic frequency and amplitude
for typical (non-head-on) impact parameters are
r '.:::::
J c - 20M 1
1 = (
10 - 4 Hz
M1
)
10 8 M 0
' (47a)
384 K. S. Thorne
(47b)
where M 1 is the mass of the large hole, M 2 is that of the infalling body, and
r0 10 Mpc might give a reasonable event rate since there are - 100 galaxies
-
as massive as or more massive than our own inside this distance, including
M87 for which observational data suggest a central black hole of mass M 1 -
4 x 109 M 0 (Section 8.3. 1 .4 of Chapter 8). These he and fc are plotted in
Fig. 9.4 for several interesting sets of parameters. It is conceivable that such
plunge bursts will be seen by beam detectors in space, if and when they are
flown.
(g) Cherished beliefs

The above discussion makes clear how uncertain is our electromagnetically
based knowledge of gravitational-wave sources. Correspondingly, it seems
very likely that when gravitational waves are finally seen, they will come
predominantly from sources we have not thought of or we have
underestimated ; and it seems quite possible that the waves will be stronger
than the above estimates suggest.
In light of this, it is interesting to ask the following question : How strong
could be the strongest bursts that strike the earth on average three times per
year without violating our 'cherished beliefs' about the laws of physics and
the universe? Zimmermann and Thorne ( 1980) have enumerated a set of
cherished beliefs , including (i) that general relativity is correct , (ii) that we do
not live in a special time or place in the universe, (iii) that there are no
enormous primordial bursts, (iv) that no single, coherently radiating object
in our galaxy has a mass exceeding 10 8 M 0, and (v) that the strongest bursts
were not beamed by their sources into solid angles � n. From these
cherished beliefs they have derived the upper limit on the 3/year burst
strength which is shown in Fig. 9.4. That limit could actually be achieved at
frequencies f <; 30 Hz by an unlikely but not implausible scenario in which a
large fraction of the mass of the universe was cycled, long ago, through a pre
galactic population of massive stars ('Population III' stars), leaving much of
the universe's mass in compact binary systems that inhabit the hal os of
galaxies. If the mean lifetime of these binaries against spiraling together and
coalescing is of order the age of the universe, then such coalescences in the
halo of our own Milky Way galaxy would give bursts in the frequency
domain f ?::, 30 Hz at the cherished belief level . For further discussion see
Zimmermann and Thorne ( 1980), and Bond and Carr ( 1984).
While this scenario is unlikely, it is not totally implausible, and it serves to

remind us that our best estimates of the waves bathing the earth could be
grossly pessimistic.
9 .4 .2 Periodic sources
(a) Characterization of the waves from periodic sources and the noise in a
Detector searching for them
The gravitational waves from a periodic source will be characterized by a
discrete set of frequencies, and the waves at a given frequency will typically
be right-hand or left-hand elliptically polarized ; i .e. for some suitable choice
of polarization axes ex' • ey' • they will have the form (similar to that of a
decaying binary, equation (42), but with constant frequency)
h + (t) h0 + cos 2nft,
= h " (t) = + h 0 " sin 2nft. (48)
If one wishes to quantify in a precise and standard manner all the properties
of these waves, including the orientations of the 'preferred' x' and y'
polarization axes, one might best do so using 'Stokes Parameters' analogous
to those used in electromagnetic theory (Section 15 of Chandrasekhar,
1950). However, in this chapter we shall be concerned only with the
frequencies f of the waves, and at a given emitted frequency, with a suitably
defined characteristic amplitude he and a corresponding noise amplitude hn
in a detector searching for the waves. ,
As an aid in defining he and h n , consider the following situation (analog of
that for burst sources in Section 9.4. l (a)) : a theorist tells us the frequency f,
the phase, and the amplitudes h 0 + ( 1 , /3, r) and h0 (z, /3, r) to be expected from
x
a specific model for a source, with orientation angles, z, f3 and distance r.

Suppose, further, that this type of source is distributed randomly
throughout the universe and that the mean number of sources inside the
distance r0 is n0 • What, then , will be (on average) the signal-to-noise ratio for
the n0th brightest source that a detector, broad-band or narrow, will see in a
search (at known frequency and phase) lasting a time i? By virtue of
equations (27) and by analogy with equation (33) for burst sources, the
answer turns out to be
(49)
Here
(50)
(with ( · · ·) denoting an average over z and /3) is the characteristic amplitude
386 K. S . Thorne
of the periodic source (analog of (3 1b) for a burst source), and
I + (e, ¢, l/l)l2>!
h (f) = < F
_[Sh (f)/f] 1 (5 1 )
"
i s the noise amplitude. (The he of equation (50) is (1)! larger than one naively
would expect from equation (27). This factor (1)1 is an approximate
correction for the fact that the angle averages in S/N should not be over
squares but, rather, over squares associated with the rotation of the earth
during data collection, then over (squares)1 covering the rest of the sky and
the orientation of the source, followed by a ! power after averaging ; cf. the
discussion preceding equation (30).) Correspondingly, if experimenters wish
to be 90 % confident of having seen that n0th brightest source after ! year of
search, then S/N must exceed 1 .655 � 1 . 7 . (Gaussian probability
distribution), and he must exceed
1.
h ,1,, = l 7 h
. 0 I
+
� 1
< F 2 )l [S,(f) x 0 - 1 Hz]l if f and phase are known .
(52a)
In the case that theory and electromagnetic observation have failed to tell us
in advance the phase and frequency of the source, except to within df, the
experimenters must try f/ilf values of the frequency, and correspondingly
,....,
the Gaussian statistics of the noise will produce
h ,,,, "" [2 ln(fINlJ lh . (f)

[2 l
��(lJ1
[s,(f) x 10 - 7 Hz] l
if f is known only to within df,...., f. (52b)

Fig . 9.6 shows he , for several postulated types of source (to be discussed in
the following subsections), and correspondingly h3/yr with f and the phase
known, for several types of detectors (to be discussed in Section 9.5 below).
(b) Rotating neutron stars

A rotating neutron star (e.g. a pulsar) will emit gravitational waves at several
frequencies as a result of deviations from symmetry around its rotation axis
(deviations from 'axisymmetry'). The larger are those deviations and the
more rapidly they rotate, the stronger will be the radiation.
Deviati ons from axisymmetry could arise in several ways : (i) The star's
solid crust (well-established) or a solid core (not so well established) could
support deformations that are residual remnants of the star's past history - a
history that might b e quite complex , including star quakes in which the crust
or core cracks and deforms in much the manner of the solid earth in an
earthquake ; for detailed discussions see, e .g. Pandharipande, Pines and
Fig. 9.6. The characteristic amplitudes he (equation (50)) and frequencies f of waves from
several postulated periodic sources (thin curves), and the sensitivities h3/yr of several existing
and planned detectors (thick curves and circles) (h J/yr is the amplitude he of the weakest source
detectable with 90 % confidence in a 1 y r = 10 7 s integration if the frequency and phase of the
source are known in advance; equation (52a)) . The sources shown in the high-frequency
region , f <: 10 Hz, are all special cases of rotating, nonaxisymmetric neutron stars (Section
9.4 .2(b)). The steeply sloping dotted lines labeled NS Rotation refer to rigidly rotating
neutron stars with moment of inertial Izz 1045 g cm 2, and with various ellipticities e and
-
distances r labeled on the lines (equation (55)) . The sources in the low-frequency region,
f ;S 0 . 1 Hz, are all binary star systems in our galaxy (Section 9.4.2(c)) : several specific, known
binaries, which are indicated by name (µ Seo , V Pup, . . . ); the strongest six spectral lines from
the famous binary pulsar PSR 1 9 1 3 + 16; and the estimated strengths of the strongest white
d ward ('WD') and neutron-star ('NS') binaries in our galaxy. The detectors are discussed in
detail in the indicated subsections of Section 9.5.
10 10
\
\
\
• ,IJ. SCO
\
I0-20 10-20
\ i Boo
\ ;,,· • · . , • • WO binary
• v Pup
• sl'IOrtest
\ brri=:::s1 Wo
period
•SS CyQ
\
4. � (r:: )
\
I
•WZ Sqe
\
)( - 1
• Am CVn
\ /
• Cy9
\ 4
/.,'?
\ s5tron9·�·,� • • • ,,
�c,'�
\ N b /_ f::J�
/rl'
._o'' .$'
shortest
�:�''•s
/"IJ
•
\ 4.2 NS binary
�riod
• \.. � · - - - - - - _ _ ./
_
PSR 1913+16 •
_
Btarn 1n Spoce 5 51b)
10- � 10·1
FreQuenc y f, Hz
388 K. S . Thorne
Smith ( 1976) and references therein. In old pulsars that have been spun up
by accretion to near-millisecond rotation rates, theory and observational
data suggest that the crust and core are quite well annealed into a nearly
axisymmetric shape (Alpar and Pines, 1985); but in neutron stars that are
only tens or hundreds or thousands of years old , it might well be otherwise
(e.g. Zimmermann, 1978). (ii) The star's internal magnetic field , if sufficiently
strong , could produce sufficient magnetic pressure to distort the star
significantly (Zimmermann, 1978 ; Gal 'tsov, Tsvetkov and Tsirulev, 1984).
However, 'sufficiently strong' means, in the case, e.g., of the Crab Pulsar, ten
times stronger than the star's measured surface field . (iii) If the star is
rotating more rapidly than a critical rotation period , Pcri t - 0.7- 1 .7 ms
(which depends on the star's structure and its temperature-dependent
viscosity) , then an instability driven by gravitational radiation reaction
('Chandrasekhar ( 1970)-Friedman-Schutz' ( 1978), or 'CFS' instability) will
create and maintain significantly strong hydrodynamic waves in the star's
surface layers and mantle, propagating in the opposite direction to the star's
rotation ; and these will radiate strongly. For detailed discussions see
Wagoner ( 1984) , Lindblom ( 1986, 1987), Friedman , Ipser and Parker
( 1986) , Schutz ( 1987), Cutler and Lindblom ( 1987).
At present we are extremely ignorant of the degree of asymmetry in
rotating neutron stars, and accordingly we are ignorant of the strengths of
the periodic waves to be expected from them . Pessimists will note that there
is no observational evidence in any observed pulsar for sufficient non
axisymmetry to produce interestingly strong waves. Pessimists will point ,
especially, to the extremely small slowdown rate of the 1 .6 ms pulsar
PSR 1937 + 2 1 , which implies such weak radiation reaction that the
characteristic amplitude at earth cannot exceed 1 x 10 - 2 7 (Fig. 9.6), and the
star's non-axisymmetric ellipticity cannot exceed 3 x 10 - 9 .
Optimists will also point to PSR 1937 + 2 1 and some other millisecond
pulsars, and note a reasonably likely scenario for their origin (van den
Heuvel , 1984; Wagoner, 1984) : that they were spun up long ago by accretion
from a binary companion until they hit the CFS instability, that they
remained just beyond the instability point for awhile, with the spinup torque
of accretion being counterbalanced by gravitational radiation reaction , and
that the accretion stopped long ago leaving the stars plenty of time to anneal
and settle down into their presently observed , highly axisymmetric states.
Given the extreme observational difficulty of finding by electromagnetic
means evidence for rapid neutron-star rotation (see, e.g. , Section IV of
Reynolds and Stinebring ( 1984) for searches in the radio), it may well be that
there are a number of accreting neutron stars in our galaxy now in the CFS
regime, radiating strong gravitational waves. For such a star the energy
being radiated in gravitational waves and that being radiated. as accretion
induced X-rays will both be proportional to the accretion rate; and
consequently the characteristic amplitude of the gravitational waves at earth
will be proportional to the square root of the X-ray flux arriving at earth, Fx :
h � 2 x 10 - 2 1
c
(300fHz)1( 10 - 8 ergFxcm - 2 s - 1 )1 (5 3)
(Wagoner, 1984). The frequency f of the waves will be f lvp /(2nR) where R
=
is the star's radius, l 3 or 4 or 5 is the spherical-harmonic order of the

=
hydrodynamic wave, and v P is the pattern speed of the wave as seen in the
inertial frame of distant observers. The X-ray flux Fx "" 10 - s erg cm - 2 s - 1
is z1o that of Seo X- 1 , the brightest quasi-steady source in the sky and itself a
candidate for a CFS-unstable object. As Fig . 9.6 shows, stars with X-ray
fluxes as low as 1 lo o Seo X- 1 could b e interestingly strong sources of
gravitational waves. In such a star the density waves in the surface layers
should modulate the emitted X-rays, but the sensitivities of past X-ray
telescopes have been too poor to detect such rapid and weak modulations.
There is an interesting proposal (Wood et al. , 1986) for a new, more sensitive
X-ray telescope designed to search for such modulations in Seo X- 1 and
other, weaker X-ray sources. Such a telescope, operated in coordination
with gravitational-wave detectors, might one day give a wealth of new
information about neutron stars.
Even the most rapidly rotating of neutron stars will be smaller than a
reduced wavelength of its emitted gravitational waves and thus can be
described with reasonable accuracy by a slow-motion formalism (Ipser,
197 1). If the emitter is CFS density waves, the multipoles involved are l 3,=
4, o r 5 . I f the emitter i s solidly supported o r magnetic-field supported

=
deformations, the waves should be largely quadrupolar, l 2. Let us focus
now on the latter case, i.e. on a star with rotation period large enough that
the CFS instability does not act. Although gravity inside the star is
significantly non-Newtonian, one can still define for the star a moment-of
inertia tensor Iik equal to the ratio of its angular momentum to its angular
velocity (both being vectors defined in the weak-gravity region well outside
the star; see Thome and Giirsel , 1983). If the star rotates about a principal
axis of this moment-of-inertia tensor, i.e. if its angular momentum and
angular velocity are parallel , then it will not precess, and (assuming a
rotation period sufficiently long that it is CFS stable), its gravitational waves
390 K . S . Thorne
will be emitted at twice the rotation frequency. From the quadrupole variant
of the slow-motion formalism we can then compute that the waves will have
the standard periodic form of equation (48) with amplitudes
- Jyy ) (nf ) 2
h o + _ 2(l + COS 2 l) Clxx
- ,
r
)(nf) 2
h 0 x
= 4 cos l (cfxx Jyy-
(54)
r
Here i is the angle between the neutron star's rotation axis and the line of
sight from the earth, and cfxx and Jyy are the components of the star's
quadrupole moment along the principal axes in its equatorial plane. The
characteristic amplitude of these waves (equation (50)) is
() i l f2
h c = 8 n 2 _3__ e zz = 7 . 7 x 1 0 - 2 08
15 r
(
I zz
)( ) (
f 2 10 kp c ,
1045 g cm 2 1 kHz r
)
(55)
where I iz is the moment of inertia of the star about its rotation axis (so
-1Izz (nf) 2 is its rotational energy) , and
8 cfx x - Jyy (56)
1z-z-
=
is its 'gravitational ellipticity' in the equatorial plane. All neutron stars for
which masses have been measured have M near l .4M 0; and depending on
the equation of state these masses correspond to 3 x 1044 g cm 2 ;S l:z:z::S 3 x
1045 g cm 2 . The likely values of the ellipticity t: are far less clear.
The observed slow-down rates of the Crab (f = 60 Hz, r = 2 kpc ), Vela
(f= 22 Hz, r = 500 pc), and PSR 1937 + 2 1 (f= 1 .25 kHz, r = 5 kpc) pulsars,
if due to gravitational radiation reaction (possible but not likely) ,
correspond to t: � 6 x 10- 4, 4 x 10- 3 , and 3 x 10- 9 respectively ; and to he �
8 x 1 0 - 2 5 , 3 x 10- 2 4, and 1 x 10- 2 7 . Zimmermann ( 1978) argues that
reasonable values for the Crab and Vela are t: - 3 x 10 - 6 and 3 x 10- 5
corresponding to h - 4 x 1 0 - 2 7 and 2 x 10 - 2 6 • Alpar and Pines ( 1985)
c
suggest reasonable values for PSR 1937 + 2 1 (which is old and well annealed)
in the range t: - 4 x 10- 1 0-l x 10- 1 1 corresponding to h c - l x 10 - 2 8 .
Blandford ( 1984) points out that, if there is a population of young pulsars
(not yet discovered) that are spinning down by gravitational radiation
reaction on a spin-down timescale -r G w, then (i) the nearest will be at a
distance r � R a{-r 8 /-r Gw )i (assumed � -r Gw ) is the mean time between
births of these pulsars in our galaxy and RG is the radius of the galaxy's disk ;
(ii) the flux of gravitational-wave energy at earth from the nearest such
pulsar (3/64n)(2ef)2h;, will be equal to [I22(nf)2/r Gw ] (4n r2) - 1 ; and (iii) as a
consequence, the characteristic amplitude of the waves from the nearest one
will be
[
he '::::'. � zz
/ Jt [
3 r r Gw
�zz
'::::'. � Jt
3 RG r B
1 . l x 10 _ 2 5(
,...,_,
)
1 04 years !
, (S7)
r8
independently of its frequency and ellipticity.
Fig. 9.6 shows values of he and f corresponding to some of the above
possibilities.
If the star does not rotate about a principal axis of its moment of inertia,
then it will precess . When one idealizes the star's interior as rigid, then
although the interior gravity is significantly non-Newtonian, the precession
is still described by the classic equations of Euler (see Thorne and Giirsel ,
198 3 , for a proof) ; and the resulting waves will have a form that typically will
entail significant spectral components at three frequencies : twice the
rotation frequency , and the rotation frequency plus and minus the
precession frequency (Zimmermann , 1980; Zimmermann and Szedenits,
1979) . In reality, pliability of the neutron-star material will cause significant
deviations from rigid rotation ; but these three frequencies may still be
dominant. If the star is idealized as a fluid body deformed by the pressure of
an off-axis internal magnetic field, then the star does not precess and the
radiation is emitted at the rotation frequency and twice the rotation
frequency (Gal'tsov, Tsvetkov and Tsirulev, 1984) .
If and when gravitational waves from rotating neutron stars are detected,
they may carry a wealth of information about the star's structure and
dynamics in the amplitudes and relative phasings of their various spectral
components. Especially interesting may be the evolution of the various
spectral components after a star quake; together with electromagnetic
timing of the post-quake rotation, these may give us new insights into the
coupling of the solid crust or core to the fluid mantle.
(c) Binary stars

Ordinary binary star systems are the most reliably understood of all sources
of gravitational waves. From the measured mass and orbital parameters of a
binary and its estimated distance, one can compute with confidence the
detail s of its waves. Unfortunately, ordinary binaries have orbital periods no
shorter than about an hour and, correspondingly, gravitational-wave
frequencies f � 10 - 3 Hz. The shortest known binary of all is a white-dwarf/
392 K . S . Thorne
neutron-star system with orbital period 1 1 minutes and gravitational-wave

frequency f � 3 x 10 - 3 Hz (Priedhorsky, Stella and White, 1986). Because
of seismic noise, detectors in earth laboratories cannot hope t9 see waves of
such low frequency. However, beam detectors in space, tentatively planned
for the turn of the century, should see them with relative ease (cf. Section
9.5.5(b) below).
Detailed formulas for the waves from a binary star, including the effects of
the eccentricity and inclination of its orbit, have been derived from the
quadrupole formalism by Peters and Mathews ( 1963) and Wahlquist ( 1987).
See also Wagoner and Will ( 1976), and Gal'tsov, Matiukhin and Petukhov
( 1980) for post-Newtonian corrections. By virtue of the eccentricity of the
orbit, waves will be emitted in equally spaced 'spectral lines' at twice the
orbital frequency and harmonics thereof. For eccentricity t: $ 0.2 the line at
J = 2forb is dominant; for t: � 0.5 the lines at f/f0rb � 2 through 8 are all
strong ; for t: � 0.7 the lines at f/f0rb � 4 through 20 are all strong . In the low
eccentricity case t: $ 0.2 the waves have the form (42) with J = 2forb which
corresponds to a characteristic amplitude (equation (50))
'
h, = 8 (i2s)' � nM
( J)l
10 - 2 1 (_L_) (�)1( ) ( )
100 pc 1,
= 8 .7 x _{ (58)
M0 M0 r 01 Hz
where µ is the reduced mass and M the total mass of the system. This
amplitude is plotted in Fig. 9.6 for a few of the most strongly radiating
known binaries. For lists of the most strongly radiating binaries and their
characteristics see Braginsky ( 1965) and Douglass and Braginsky ( 1979).
White-dwarf and neutron-star binaries should also be important emitters
- and they should extend to higher frequencies than ordinary binaries ; but
there is a paucity of observational data on them and the example of shortest
known period has f only 3 x 1 0 3 Hz (see above). From the data that do
-
exist (e.g . lben and Tutukov, 1984), Lipunov and Postnov ( 1986), Lipunov,
Postnov and Prokhorov ( 1987), and Hils et al. ( 1987) have estimated the
characteristic amplitudes of the strongest white-dwarf and neutron-star
binaries ; see Fig. 9.6. For a very detailed treatment of the white-dwarf case
see Evans, lben and Smarr ( 1987). The highest frequency to be expected for
any white-dwarf binary in ou r galaxy is 0.06 Hz since mass transfer from the
less massive star to the more massive begins at or before this frequency; the
highest for any neutron-star binary is 0.007 Hz since any binary of higher
frequency than this would coalesce in a time less than the mean interval
between coalescences, "' 10 4 years.
Gravitational
evolution of close radiation
binary reaction
systems; seeplays an
Paczynski important
and rol e
Sienkiewicz i n dri
( v i n g
198 1) for
the
details. 9.4 .3 Stochastic sources
(a) Characterization of stochastic gravitational waves
Ittraveling-wave
is useful to normalthink about modes stochastic
of the gravitati
gravi t o
ationalnal waves
fi e ld. i
As n terms
wi t h of
the
electromagneti
states) for each c field,
vol u me there
(2nh) are two modes (because of
3 in phase space; and correspondingly, one can two pol a ri z ati o n
easily show, the energy densi t y per unit logari t hmic interval of frequency
.universebyi.s the critical energy density Perit "" 10- 8 erg cm- 3 to close the
divided
n (f)
d
=
E GW/d 3x d n f ___!!____ ( f )3
I _
(59)
Gw - 1037 1 kHz
Pent.
Here ii is the average number of quanta i n al l modes with frequenci e s of
order f. Below we shall see that n Gw( f) is likely to be ;;;:: 10-14 at all
frequencies of interest, and correspondi n gly the mean number of quanta i n
each mode is likely to be ;(: 1020 - so large that a classical treatment is in
order.
In TT coordinates the metric perturbation due to stochastic gravitational
waves, evaluated at any chosen location xi, will be a sum over contributions
of all the modes of the field
TT
h jk ( t, ) X
i
=� TT
h Kjk ( t, )
L..- X
i
•
(60)
K
Here the i n dex K labels modes of the field. For stochastic gravi t ati onal
waves, the wave field hljk associ a ted with mode K can be regarded as a
'random process' (i.e. a stochastically fluctuating function of time t) ; and the
total field h°j,[ at location xi is the sum over all of the modes' random
processes.
The field of a chosen mode K can be expressed
TT .
h Kjk h K (t, x1)eik '
=
K
as
(61)
where h K (t) is its scalar wave function and e� is it s constant polarization
vector, so normalized that e�e� 2 i n Cartesian coordin ates; cf. equation
=
(7c). Then h K is a scalar random process in time (at fixed xi) and its statistical
properties Can be characterized by a Spectral density shK (f) .
I shall assume that the modes are defined i n such
significant correlation between their wave fields hVjk . As a result, when one a way that there i s no
averages the stress-energy tensor (9) of the waves at xi over a sufficiently
394 K. S . Thorne
long time, al l cross-terms between diffe rent

obtains for the time-averaged specific intensity I at location modes get washed x
outi
and one
1
i
. dEd d df! = .L 4 nf 2
I (t )
, x
•
d f! = (62)
1 dA d t f Q K m �n ShK ·
Here E denotes energy, A denotes area, n denotes solid angle, and the sum is
over all modes K with propagation directions in the infinitesimal solid angle
.1!1. The total energy per unit logarithmic i n terval (used in defining !1 ( f ) 0 w
nf3
above) can be expressed in terms of the specific intensity in the standard way
dE = fl dQ = t 4
f! aw (f) P erit =
d d ln f I
3
X
J Sh K , (63 )
where
As the
for integral
burst and i s over
periodic the entire
waves, so sphere
also and
for the sum
stochastic, i
we s over
shall al l modes.
introduce
asituation:
single characteri s ti c amplitude he that is tied to a specific experimental
the experimenters use two identical gravity-wave receivers
(broad-band or narrow), separated by a distance 4: l. = c/2nf, to search for
isotropic, stochastic waves in the neighborhood of frequency f. The search is
performed by a standard technique (Bendat, Section of Drever, 1958 ; 9
1983 ) : 1
the outputs, h 1 (t) and h 2(t) of detectors and 2 are passed through
i.1denti� calcentered
filter� onwhifrequencies
ch admit onlyf. The Fourier
fi l components
tered outputs in a bandwidth
w 1 (t) and w 2 (t) are
f f +
then multiplied together and integrated for a time tof get a si n gle number
W= n�+i W 1 (t)w 2 (t). This number will consist of a signal due to the identical
stochastic backgrounds contained in w 1 (t) and w 2 (t), and a Gaussian noise
due to the independent noises i n the two detectors
7
(each with
1958 )
the same
spectral densi t y Sh (f)). It turns out (e. g . Chapter of Bendat that the
ratio of the signal to the root-mean-square noise is
(64)
where
h, ( f) = [t JS•/f) T [� f dQT = [";2
= T
11 ila w U )P, n1
= 1 .3 x w- is( 8 erg cm )1( 1 Hz ) [Q

1 . 7 x 10 -
Pe ri t
- 3
f
Gw
(/)Ji
(65 )
is the characteristic amplitude of the isotropic, stochastic waves, and

1 [fSh (f)] 1
h (f) _
(66)
n - (tfdf)i (F! ) i
is the characteristic noise amplitude of the detectors. (Note: P e ri t/ 1 .7 x
10-8 erg cm- 3 =(H0/lOO km s- 1 Mpc- 1 )2 where H0 is the Hubble
constant. ) Correspondingly, i f the experi m enters wish to be 90 % confident
of having seen the stochastic background during a search of duration =� i
year, the Gaussian probability distribution for S/ N requires S/ N = 1. 7 ,
he
which in turn means that must exceed the noise level
h (f )= l . � (f )= 2 .0 ( _!lf )-� [f h f)Ji
S( i
3/ yr hn (67)
10 7 Hz (F i ) .
The characteristic an:iplitudes he of various possible stochastic sources,
h 3;yr
and the noise levels of various detectors are shown in Fig. 9. 7 .
(b ) Binary stars
So many binary stars in our galaxy and in other galaxies radiate in the
frequency region ;S
f 0. 03 Hz that they should superpose to produce a strong
stochastic
and Prokhorov background.
(1987) and Lipunov
Hils and Postnov (1986), Lipunov,
et al. (1987) have made careful calculations
Postnov
of the characteristic amplitude of this stochastic background as a function of
frequency; the results of Hi l s et al. are shown in Fig. 9. 7 for the contribution
ofcontributions
our own galaxy of (which
all other should
galaxies be concentrated
(which should in bethe galactic
isotropi c )pl a ne).
should Thebe
down from those of our own galaxy by (hc )o t her/(hc )u s - 0 . 1 5 .
The binary stochastic background i n
contributions from various types of binaries. Those shown as solid curvesFi g . 9. 7 is broken up into
(unevolved binaries, WUMa stars (first discussed by Mironovskii, 1966),
and
firmly catacl
based y smi
on c variables
opti c al studi (white-dwarf
e s of the / normal-star
stati s ti c s of these systems))
types of are
stars,rather
and
thus
Evans, are rather
Iben and rel i a ble.
Smarr, Those
1987, shown
and dashed
the above (cl o se white-dwarf
references), and binari e
neutron-stars (see
binaries) are based on so little observational data and so much theory that
they are highly uncertain.
The binary background presents a serious potential obstacle to searches
forspace-based
other kindsbeamof waves i n the frequency band 0. 0 3 Hz ;S ;S
f 10-5 Hz where
above
that a thi
lM s background
0 star
detectors iwif itl hasoperate. A broad-band whiburstch can
falling onl
into y a h e burst > h e backgrou nd
supermassive black hole -in the Virgo
be seen
means, e.g . ,
cluster
will not be discernible unless the hole' s mass is M 0
< 3 105 M (cf. Figs. 9. 4
x
has
.7 ) .
( f )-i o -
and 9 A periodic source can be seen, after an integration time only if it
h e pe ri odic > i hc backgr u nd which means,
dwarf binaries are as numerous as estimated, the binaries 10z Boo e. g . , that if the
i,
close
and whi
SS Cygt e
will be discernible only after integration times of i> 7 s. For further
discussion see Evans, Iben and Smarr (1987) and Hils et al. (1987).
396 K . S. Thorne
Fig. 9.7. The characteristic amplitudes he (equation (65)) and frequencies f

of waves from several postulated stochastic sources (thin curves), and the
sensitivities h31yr of several existing and planned detectors (thick c � rves and
circles) (h3/yr is the amplitude of the weakest source that can be detected with
90% confidence in a 1 yr = 10 7 s integration). The very thin diagonal lines
indicate values of he corresponding to constant nGwU) = (gravity wave
energy m bandwidth !if= /)/(energy to close the universe if
H0 = 100 km s - 1 Mpc - 1 ). The sources shown as solid curves (background
from various types of binary stars in our galaxy; Section 9.4.3(c)) are rather
firm predictions. The sources shown as dashed and dotted curves are much
less finn; see Sections 9.4.3(c)--{f) for discussion of specific sources. Not
shown are primordial waves from the big-bang (Section 4.3(d)), which could
have nGw as large as 1 or as small as 10 - 1 4 or less according to various
plausible scenarios - but which are limited by observations as discussed in
Section 9 .5.6. The detectors are discussed m detail m the indicated
subsections of Section 9.5.
10
Sun's Normal Modes, 5.5(d)
v 5.5(0 )
Po'51 Doppler.
1986 Seams
5./7\
Frequency f , Hz
(c) Population stars Ill

Ifstars';
thereCarr, was 1986),
a pre-galacti
the violentc population
events that of
termimassin ated ve stars
their ('
livesP opulati o
(supernovae n III
and collapse to black holes) might have produced gravitationat"waves that
westarstoday
are a would
firm see
source asofi s otropi
stochasti c and
c stochasti
background, c . Whereas
these Populati exi s ting
o n bi
III n ary
stars
are a highly speculative one. Carr ( 1980) has derived an upper limit on the
characteri
such stars s ti c
under amplitude
any that
reasonable coul d have
scenari o been
; i t i sproduced
shown i nby Fi the
g . deaths
9.7 (upper
of
short-dashed curve) along with the maximum characteristic amplitude that
coul
binariesd comethat from
decay theby remnants
radi a tion of
reactithese
o n stars
on a i
tif mthey
escalebecame
-rG w �
black-hole
10 1 0 years
(Bond and Carr, 1984).
(d) Primordial gravitational waves
Photons comi n g from z "'
the big bang
cosmological redshift 1000 when the universe0was roughly one million last scattered off matter at a
years old; and neutrinos last scattered at z- 101 when it was about 0. 1 s
old. An order-of-magnitude calculation shows
last scattered at roughly the Planck time, i.e. during the first 10- seconds that gravitons, by 4
contrast,
3
when
diunlikely spaceti
fferent from m e was
today quanti
(Secti o z
n ed and the laws of physi
7.2 of Zel'dovich and Novikov, 1983). (An
c s were exceedi n gly
exception occurs if, at the epoch when the waves' reduced
wavelength was ..t - (horizon size), much of the uni v erse' s energy densi t y was
iabsorpti
n relativoisntimayc partioccur;
cles wiseeth Vishni
mean free paths of order ..t; then non-negli g i b le
a c ( 1982).) Thus, i n �tudying primordial
gravitational waves (waves created in the big-bang), one usually can ignore
theirNotsubsequent
so for theiri n teractions
subsequent with matter.
interactions with the background spaceti m e
curvature of the universe. Gri s hchuk ( 1974, 1975a,b, 1977) has shown that,
as the primordial perturbations that give rise to present-day waves 'come
inside the cosmologi c al horizon' -and also before
they can be parametrically amplified by their interaction with the dynamical they enter the hori z on -
background spaceti m e curvature; i n other
graviton creation. In this way exceedingly small initial fluctuations can bewords, they can tri g ger fu rther
ampliJust fi e d
how i n to
muchan interestingly
stochasti c strong
background stochasti
is c
produced background
depends today.
cruci a lly on
ill-understood aspects of the initial singularity and on the equation-of-state-
398 K. S. Thorne
dependent and
Some otherwise vacuum-dependent
plausible models expansion
produce so rate
much, i n the very early universe.
n0w(f ) � 1, as to be in
violent
Zel' d conflict
ovich and with the observed
Novikov, 1983). current
Other, state of the
equally universe
plausible (e. g .
models p. 621 of
can
produce so little, now(f ) � 1o-14, that there is no hope of detecting the
waves in the foreseeable future.
In currentl y fashionable inflationary models of the
fluctuations which initially are smaller than the horizon (A: � 9f8by universe vacuum =
(background radius of curvature)) are dri v en outsi d e the horizon (A: � Bf8)
the inflationary
constant expansion.
amplitude While outside the horizon, they are '
h. Much later, after inflation ends, non-inflationary
fr ozen' with
expansion brings them back insi d e the horizon.
each mode before entering and after leaving the horizon is The number of quanta in
n- [ 1�11 (ff}Ml'( h�x)- : ( h�) 2'
where
occupi the
e d first
by factor
each is
mode, the
and waves'
the energy
third is density,
1/(energy the
of second
one is the
graviton). volume
Before
entering the horizon n �!; so the above relation says that upon leaving it
A:leave 1 alea ve
nout � nen ter - n ter 2 a en ter ,
A:e- --
where
freezi na is the expansion factor of the universe. Thus, the epoch of amplitude
g is actually an epoch of parametric amplification (stimulated
creati
depends o n of
on new
the gravitons);
total amount and
of the
expansion total number
that occurs of gravitons
while the created
waves are
outside the horizon. (There will be additional parametric amplification as
the waves emerge from the horizon, A:,_ 9f8, but in inflationary models that is
generall y smal l compared to the ampli fi cati o n during freezing, A:� Bf 8.) The
total amount of inflationary expansion differs from one inflationary model
tounianother; and correspondingly, the models can give
ty or now too small ( � 10-14) for there to be hope of detecting the waves. now as large as
For
universe discussions
on the of
spectrumthe influence
of the of
amplified the equation
waves, see of state
Grishchuk i n the
(1977) early
and
Fig. 4 of Grishchuk and Polnarev (1980). For calculations of the waves
produced by specific inflationary scenarios see Starobinsky (1979),
Rubakov, Sazhin and Veryaskin (1982), Abbott and Wise (1984), Halliwell
and Hawking (1985), Mi j ic, Morris and Suen (1986)
Because the range of possible strengths of primordial waves is so great, weand references therein.
do not bother to show it in Fig . 9.7 -aside from indicating the values of he
corresponding to various values of naw(f) .
(e) Phase transitions
During the early expansion of the universe, there may have been first-order
phase transitions
interactions. In associated
each of these with
phase QCD interactions
transitions the and
ori g i n with
al phase El e ctroweak
woul d be
supercooled, by the cosmological expansion, below the equilibrium
temperature of the new phase. Bubbles of the new
at isolated locations and expand at near-light velocity until they have phase would then nucleate
compressed
equi l ibri u m. the
As original
Witten phase
(1984) enough
has for
pointed the
out,two
and phases
Hogan to coexist·
(1986) hasin
nalyzed in detail, this 'cavitation' should have produced gravitational
awaves
sound i n
wavestwo ways:
they (i ) directly
generate, and from
(ii) expandi
subsequently n g bubbles
from and
the the subsequent
inhomogeneities
associ a ted
inhomogeneities with the
and two
corresponding co-existing phases
inhomogeneities (l arge-scale
in the density
Hubble
expansion rate). The resulting gravitational waves should possess a
spectrum
when the that peaks
cavitation at wavelengths
occurred. Those whi c h
wavelengths were of order
correspond the to hori z on
frequenci si z
e e
s
today fmax - (2 10 - 7 Hz)(kT/ 1 GeV), where T is the temperature of the
x
phase transition. Hogan' s (1986) predi c ted spectra, shown in Fi g . 9. 7 , thus
peak at fmax "' 2 10 - s Hz (QCD, T- 100 MeV) and fmax "" 2 10 - s Hz
x x
(Electroweak, T- 100 Gev). The J+i shape of the spectra at frequenci e s

f fmax is a firm prediction, but the shape f - 1 at f < fmax is not (it could well
>
- with > 1). The amplitude shown is a reasonable upper limit, unless
P
bethefphase p
transition is unusually catastrophic with very strong supercooling.
(f) Cosmic strings
Long before
initigrand-unified the QCD
al singularity -there andmay Electroweak
have been phase
a phase transitions
transition - i. e
associ . nearer
a ted the
wi t h
the interactions, and that transition may have created cosmic
strings -one-dimensional ' d efects' in the
n vacuum
estimated to be µ- 10 - 6 and with te sion equal to mass per unit length with mass per unit length
(Zel' d ovi c h, 1980; Vilenki n , 198 la). As the universe's horizon expands to
uncover the stochastic inhomogeneities in a string's shape, those
inhomogeneities should begin to vi b rate with speeds up to the speed oflight.
Bycouldself-intersection of the stri n g, closed loops should
have acted as seeds for the condensation of galaxies and galaxy form; and those loops
400 K. S. Thorne
clusters (Zel 'dovich, 1980; Vilenkin, 1981a; Turok and Brandenberger,

1986; Sato, 1987).
An unavoidable byproduct of this model for
amounts of stochastic gravitational waves produced by the vibrations of the galaxy formation is huge
clVilenkin enkin, 198 l b).earlier,
osed loops(1985)(Vil(confirming Detailed less cal c ulations
accurate cal by
c Vachaspati
ulations by and
many
others) predi c t that, i f the strings are not superconducting (f or the
( 1 6)! for all f 10 - s Hz (-lo-6µ-) . (68)
superconducting case see Ostriker et al., 1986),
now(f) - 10 - 7 ;_ ;c;
(For the spectrum at lower frequenci e s see Hogan

less than 10 - 6 (i.e. if gravity-wave observations constrain now(/) to be and Rees, 1984. ) If µ is far
� 10- ) then the non-superconducting cosmic-string theory of galaxy
7 ,
formation
(68). From will
that face severe
diagram anddi ffi
the culties. Fig.
corresponding 9. 7 shows the
discussions predi
in c ted
Section waves
9. 5 it
isplaciclearng that several
cosmi c di
string ffe rent
theory observational
i n jeopardy techniques
- or, have
hopefully, the
of prospect
discovering of
stri5°-scale
ng-produced waves. (The apparent disproof of now "' 10 - 7 coming from
anisotropy of the cosmic microwave radiation
9. 5 . 6 (c)) does not in fact constrain cosmic strings, since this observational(Fig. 9. 7 and Section
limit is sensitive only to waves that were
wavelength)-(horizon size) at the epoch of recombination - before the present and had (reduced
strings that produce this wavelength began to vibrate and radiate. )
9.5 Detection of gravitational waves
9 . 5 . 1 Methods of analyzing gravitational wave detectors
When analyzing the performance of a gravitational wave detector, it is
important to pay attention to the size L of the detector compared with a
..t
reduced wavelength of the waves it seeks.
If L � Jc then the detector can be contained entirely in the proper reference
frame of its center, and the analysis can be performed using non-relativistic
concepts augmented by the quadrupolar gravity-wave force field (3), (5). If
one prefers, of course, one instead can analyze the detector in TT
coordinates
The two using
analyses general
are relativistic
guaranteed concepts
to give and
the the
same spacetime metric
predictions for (8).
the
detector's performance, unless errors are made. However, errors are much
more likely in the TT analysis than in the proper-reference-frame analysis,
because our physical intuition about how experimental apparatus behaves is
proper-reference-frame
example, we intuitively based
assume rather
that i f than
a mi c TT-coordi
rowave cavi ntate
y i s based.
ri g i d , i t As
s an
walls
will reside at fixed coordinate locations xi. This remains true in the
detector' s proper reference frame (asi d e from fractional changes of order
(L2/J:.2 ) h, which are truly negligible if the detector is small and which the
proper-reference-frame
there the coordi n ate anal y
locationssi s ignores).
of a ri g i But
d it
wall is not
are true
disturbedi n TT bycoordi
fracti n ates;
o nal
amounts of order h, which are crucial to analyses of microwave-cavity-based
gravity wave detectors.
Thus, for small detectors, L� J:., the proper-reference-frame analysis is
much
For to
l a be
rge preferred.
detectors, L � J:., one cannot introduce a proper reference frame
that covers the entire detector. Such detectors can only be analyzed using
general relativistic concepts in TT coordinates
other suitable coordinate system (rarely as good). (usual l y the best) or in some
9 .5 .2 Resonant bar detectors
More effort
gravity-wave has been
detector. put
Weber' into s resonant
original bars
detectors than into
were ofany
the other type
resonant-bar of
type;
were al
of l but
thi s one
type; of the
and eifirst-generation
g ht of the world' (pre-1977)
s twelve earth-based
research detectors
groups now
building
current and
bar operating
eff o rts three earth-based
are i n the Unidetectors
t ed States are working
(the with
University bars.
of Maryl Of athe
nd
(Weber, 1986), Stanford University (Boughn et al., 1982; Michelson, 1983)
and Louisiana State University (Hami l ton et al., 1986)); two are in Europe
(the4)University of Rome with its detector si t ed at CERN (Amal d i et al.,
198 and Moscow University (Braginsky, 1983)); and three are in the Far
East (The University of Western Australia in Perth (Blair, 1983), Tokyo
University (Owa et al., 1986) and Guangzhou, China (Hu et al., 1986)). The
improvements
been a factor in
of resonant-bar
roughly 200 sensitivities
i n amplitude, si n ce Weber' s
correspondingfirst detector
to have
40 000 in
energy; and significant further improvements are yet to come.
(a) How a resonant-bar detector works
Schematically
solid bar whose (Fig. 9.
mechani 8 ), caal resonant-bar
oscillations detector
are driven consists
by of
gravitatia olarge,
nal heavy,
waves, a
transducer that converts i n formation about the bar 's oscillations into an
electrical signal, an amplifier for the electrical signal, and a recording
402 K . S. Thorne
system.
sensor. The transducer and amplifier together are sometimes called the
The transducer typically is mounted on one end of the bar.(though other
mountings are sometimes used), and it produces
current proportional to the displacement x(t) of the bar's end from an output voltage or
equilibrium. Although x(t) is a sum of contributions from all the "" 10 2 9
normal
sopassed modes of the bar, the transducer' s output is
that only the contribution of the bar's fundamental normal mode is filtered by the amplifier
on through. This is accomplished by a band-pass
the frequency fo of the fundamental mode, with bandwidth 13.f somewhat filter centered on
smaller than the diffe rence - 0
f1 f between the bar' s fundamental and its
first harmonic.
asirreltheevant. Thus, in effect, it is the fundamental mode
gravity-wave detector; and all the other normal modes are almost of the bar that acts
Si n ce the fundamental mode involves the relati
bar's left and right ends with just one node (at the bar's center), itv e in and out motion of the
corresponds to a standi n g sound
the bar. Correspondingly, the bar's length must be wave with wavelength twice the length of
(69 )
where vs i s the speed of sound i n the bar. Typi c al solid materials have
longitudin1 al sound speeds of order 5 km s - 1; astrophysics suggests (Section
9.4) that kHz is a reasonable frequency to search for gravitational waves;
and correspondingly the lengths of typical resonant bar detectors are about
2 m and their masses are several tonnes. Notice that equation (69 ) gives for
the ratio of the length of the bar to the reduced wavelength of the
Fig. 9.8. Schematic diagram of a resonant-bar detector for gravitational
waves. The angles (9, </>, l/I) characterizing the propagation and polarization
directions of the waves relative to the detector are a specialization of the
angles (8, </>, l/J) shown in Fig. 9.2.
S ENSOR
{ TRANS DUCER
AMPL I FIER
)(
gravitational waves, A = c/2nf0 ,

L/J: ::= nv5/c ::= 5 x 10 - s . (70)
Thi s justifies,
reference-frame with hi g h
viewpoint accuracy,
i n the
analyses use
of of a
bar Newtonian-language,
detectors. I shal l adoptproper
that
viewpoint i n thi s chapter.
The contribution of the fundamental mode to the displacement x(t) of the
bar's end can be expressed in the standardi harmonic-oscillator form
x(t) = Real [X ( t) e - 2 nf01 J , (7 1)
where (72)
isthethequantity
mode'sXcomplex amplitude and fo is its eigenfrequency. It is actually
(t) that the sensor monitors. To the extent that one can ignore
forces from
toX(t).theWhen the
other modes), fundamental
onl y mode'
gravitational s environment
waves will (e. g .
produce very weak
ti m e coupli
changes n
ofg
waves hit, they drive X (t) to evolve in a complicated, time
dependent
someUnfortunately, way, and
details of thenoise that
waveformevolution i n
h j,[ (t) .
pri n ci p le can be deconvolved to reveal
in the sensor is typically so severe that to control it
the experimenter must use a bandwidth !lf that is far smaller than the
frequency fo (typically !if/f0 - 0.0 1 in present detectors). Correspondingly,
the sensor averages X (t) for a time f ::= 1/!lf that is long compared toXthe
peri o d P 0 = 1/f0 of the gravitational waves being sought, before passing (t)
onburststo the(e.grecording
. from system.
supernovae), Thi s
whichmeans havethat for
durationstypi c
ofal gravitational-wave
only a few ti m es P0,
allfromthatbefore
can bethemonitored is the total change /l X in the complex amplitude
wave arrives until after it has passed. Only uncommonly
long bursts, those lasting for more than f0 /llf - 100 cycles, can be
monitored i n greater detail. In the future, however,
the sensor noise under better control and thereby opening up the bandwidth there i s hope of bringing
!lf ::= 0.2f0 to permit detailed monitoring of much shorter bursts
to(Michelson and Taber, 1984) . (For a description of several first-generation
broad-band bars see Figure 2 of Drever, 1977 and associated text, and
references therein.)
(b) The sensitivity of bar detectors to short bursts
A bar detector couples to the field (equation .(26))
h(t) = F + ( O , </> , t/!)h + (t; {3) + F x ( O , </> , t/!)h x (t {3),
l, ; l, (73)
where, if the bar is axially symmetric and the direction (0, </>) and
404 K. S . Thorne
polarization angle t/J are defined as in Fig . 9. 8 , the beam-pattern factors are
+
F = sin2 cos 21/1 ,
8 = F " sin 2 sin 2t/J.
8 (74)
(See Chapter 37 of MTW for the key elements of a derivation. ) We shall
presume, throughout this subsection, that h(t) is a burst of such short
i
duration L\t ;::::; = 1/L\f that the optimal way to search for it is to measure the
mean square change IL'.\ X l 2 it produces in the fundamental mode ' s complex
amplitude. In this case the general formul a (29) for the rati o S 2IN 2 of the
burst's squared signal to the mean-square Gaussian noise in the detector can
be1981;reduced to the simple
Michelson and Taber, 1984) form (see, e. g . , Giff a rd, 1976; Pallotino and Pizella,
S 2 tMerr(2nfo ) IL'.\ Xl 2 (75)
Here k is Boltzmann ' s constant, Tn is a ' n oise temperature' which
characteri z es the overall noise in the detector, and M err is an 'effective mass '
associ a ted with the fundamental mode, so defi n ed that !M errlXl2 (2 nf0 ) 2 is
the(Sintotal
ce X energy
is actual inl the
y themodeamplitudewhen it
of is vi b
motion rati n
ofg with
the endcomplex
of the amplitude
bar, for a X.
bar
that has uniform cross-section and is long compared to its diameter, the
effective mass is Merr = 1( 1 + v2)M where v is the Poisson ratio of the bar's
materi a l and M is the bar' s mass. )
iequatiBecause
ndependent the
of net
the wave-i
mode n duced change L\X in complex
's initial complex amplitude, the numerator of amplitude is
on (75) is the energy that the wave would have deposited in the mode if
the mode had been initially unexcited. This deposited energy can
conveniently be expressed in terms of the cross-section (J 0( f) that the mode
would present to the wave i f the wave had hit
(broadside, = n/2) and with an optimal polarization ( + mode with t/J = 0):
e
it from an optimal direction
1N ,rr(2 nfol2l,U I 2 � r � f 2 lii( Jll 2a " (fl df. (7 6)
Here h( f) is the Fourier transform of h(t) (equation (73)), and for an optimal
direction and polarization (n/2)f2 l fi( J)j 2 would be the energy per unit area
per unit frequency ( f 0) carried by the waves. Because 0( f) is extremely
� (J
sharply peaked around the resonant frequency fo (Section 37. 5 of MTW),

we can rewrite (76) and thence (75) in the form
�: = ; fo2l fi(fo ) l2 J ��df . n
(77)
This is a special narrow-band-detector version of the general equation (29)
for arbitrary detectors. Correspondingly, equation (30) for the strongest

burst seen, on average,
our source, becomes at the same rate as bursts occur inside the distance to
J <Jo df
� -
= �
N2 2 2 f 2
0n( I -
h + ( fo ) 1 2 + 1 -
h x ( fo ) 1 2 ) (F 2
+ )
kT
, (78)
n
where, for the F + and F " of (74),

4
< F 2+ ) = (F 2x ) = i5 = 0.267 . (79)
Equation (78) motivates us to define
h e = ( 3 )1fo < lii+ ( fo)l2 + Iii" ( fo)l 2 >1 (80)
the characteristic
asdetectors) amplitude of the burst (analog of (3 lb) for broad-band
and
(8 1)
asbandthe detector's characteristic noise amplitude (analog of (32) for a broad
detector). Correspondingly, the characteri s ti c ampl i t ude he = h 3 /yr that
a source must have to be detectable with 90 % confidence in a search lasting 1
year is
(82)
For
frequency a broad-band burst that is peaked near the detector' s resonant
f0 (so the fe of equation ( 3 l a) is approximately f0 ) and that lasts
a time not(80)much
foramplitude longer than flt = 1/f0 , the narrow-band characteristic
wi l be roughl y equal to the broad-band characteri
amplitude (3 lb). For such bursts, and only for such bursts, it makes sense to s ti c
plbroad-band
ot a bar detector' s narrow-band h 3 1 yr on the same graph (Fig . 9.4) as a
detector's h 3/yr· Inspiraling binaries do not belong to this class
of9.4.bursts, so their detection by bars must be discussed separately from Fig.
(c) How to optimize the sensitivities of bar detectors
Equation (82) shows that to optimize the sensitivity of a bar detector to
short, broad-band bursts, one must achieve the largest possible frequency-
406 K. S. Thorne
iTntegrated cross secti o n J a 0 df and the smallest possi b le noi s e temperature

n.
The i n tegrated cross secti o n J a 0 df can be
response of the bar's fundamental mode to the force field (5) produced by a computed by. analyzi n g the
usoidalng wave
siinntegrati over with
the wave' opti s m al direction
frequency. See and
MTW polari
Box z ation,
37. and by then
4 for details. For a
cylindrical bar with length L somewhat greater than radius R (the usual
siveltuation), the result depends only on the bar' s
E/p)i with £ =(Young' s modulus) and p=(density)(Paik and
mass M and internal sound
o ci tvs
y =(
Wagoner, 1976):
8- GMv3 2 [ 1 + 1v( l - 2v)(nR/L) 2 + .. ·J
I a0d f =7t c
s
= 1.6 10 - 21 cm2 Hz ( 10M3 kg ) (5kms- 1 )1 (83)

x
vs
(Here vis the Poisson ratio of the ) bar'

Thus,s materi
it is a l, and
desirable only
to the
build leadi n g
detectorsshape
that
dependent
areTheas massi corrections
v e and ha,. are
v e given.
as hi g h sound speed as possi b le.
noise temperature T0 is determined by a combi n ati o n of noise in the
sensorThe and thermal noise in the bar.
thermal noise in the bar i s caused by weak coupling of the bar' s
fundamental mode to its environment -its 1029 other modes, the wire or
,...,,
cabl
chamber e or prongs
(Bragi nthat
sky, suspend
Mitrofanov the bar,
and and
Panov, the resi d
1985). ual
If, gas
as in
is the vacuum
normal, thi s
envi r onment is thermali z ed at some physi c al temperature Tb (subscript ' b '
for bar or for thermal bath), then these couplings cause the mode's
amplitude to execute a random walk (Brownian motion) in the domain
:!S
IXI X th corresponding to an energy k Tb :
(84)
The fluctuation-dissi pationchanges statesXthatisthethetisame
theoremof order mescaleas onthewhich
timescal thise
random
*
walk produces th
r = Q/nf for large-amplitude vibrations to be damped fri c tionally. (Here Q
0
is the quality factor of the mode's large-amplitude oscillations: the number

radians of oscillati
ofConsequently, o n requi r ed for the energy to damp by a factor e.)
the mean square Brownian change in the mode's amplitude
during the sensor's averaging time i= 1/!!if is

f
ll!J.X th 1 2 = x2t h i_ = X2th � o
Q !!if
(85)
r*
corresponding to
i
k Tb fo
2 M eff(2 Tr.J{'o ) 2 1 l!i X , h 1 2 = TC - -.
Q !!if
.
(86)
The detai l s of the sensor noise depend on a variety of factors, including:

thestructures
detailedseestructure
Secti o n of the transducer (for a revi e w of many
4. 1 . 5 of Amaldi and Pizella, 1979), the strength of
transducer
coupling of the transducer to the bar (characterized by a dimensionless
coupling
oscillation constant
required f3
for that
al l is
the roughly
fundamentalequalmodeto the number of cycles
's energy to be fed into the of
transducer);
amplifier; the the impedance
noise temperature mismatch between the transducer
Ta of the amplifier (converted to an
and the
equivalent noise temperature (f0/fa) Ta at the bar's frequency fo if the
ampli f i e r operates at a di ff erent frequency fa than the bar); and the sensor
bandwidth !!if. Although the details vary from one sensor to another (see,
e.g ., Weiss, 1978 ; Pallotino and Pizella, 1980; Blair, 198 3 ; Michelson and
Taber, 1984), the spirit of the details is captured by the following
approximate expression for the mean square change ldX senso rl2 that the
experipresent,
menter would (ii) infer
back-action from the
forcessensor'
of thes output
sensor i
on f (i)
theonly
barthe sensor
were noise
negli g i b le
were
(see subsection (f) below), and (iii) impedances were properly matched:
t M eff( 2TCJ. f'O )2 1 L1A X sensor 1 2 � k Ta(fo /fa) !!if
2 (87)
{3 f0 ·
The sum of 1M err(2nf0)2jl!J.X senso rl2 (equation 87)) and 1M err(2 nf0)2jl!J.X ihl 2
(equation (86 )) is the mean square noise N2 that appears in the denominator
ofsum.equati o n (75); i. e . , by definition, the detector' s noise energy k Tn is this
If the experimenters choose too large a bandwidth df, then the sensor
noise (87) wil become inordinately large, producing a large Tn and masking
the gravity-wave signal. If they choose too small a bandwidth, then the bar's
thermal noise (86) will become inordinately large. There is an optimal
bandwidth, typically df"" 10 Hz in present detectors (corresponding to the
averagi n g time i --0. 1 s) at which roughly half the noise is from the sensor
and half is from thermal motion of the bar. When the bandwidth is chosen
optimally these two noise sources together produce a detector noise
408 K . S. Thorne
temperature
(88)
In practi c e, the experi m enters choose a bar earl y in their experiments,
thereby fixing the integrated cross section J a0 df; and they then struggle for
many years to develop a sensor and its coupling
cold environment, that will minimize the noise temperature Tn . Maximizing to the bar, and a thermally
J a0 df is achieved by maximizing the bar' s mass and its velocity of sound
(and, to a small extent, optimizi n g its shape
constraints such as available cryostats.) Minimizing Tn is achieved subject to other experimental
according
quality factor to equati o n (88) by (i) maximizing the fundamental
Q (i. e . minimizing its coupling to the rest of the world), (ii)
mode' s
cooling the bar to as low a physi c al temperature Tb as possible, (iii)
maxi m i z i n g the strength f3 of coupling of the
an amplifier with as low a 'noise number' kTa /(2nhfa) as possible (the transducer to the bar, (iv) using
Heisenberg uncertainty principle limits the noise number to be � 1; Weber,
1959;
matching Heff n
of er,
the 1962; Caves,
transducer 1982),
and and
amplifier (v) (astruggling
requi r to
ement get
for good
equation impedance
(88) to
be valid).
(d) Parameters of first- and second-generation bar detectors
The first-generation bar detectors (pre-1977) were aland l made of aluminum,
Q ,..., 105; they all
weighed roughly 1. 5 tonnes, and had f 1. 6 kHz
0 -::::::.
operated at room temperature, Tb - 300 K ; and most used piezo-electric
transducers
when squeezed. - i. e .
For crystals
most glued
the to
coupl the
i n g bar
of which
the produce
transducer small
to the voltages
bar was
weak, /3 � 10 - 4; but those in Britain achieved strong coupling, f3 -:::::. 0.2 and
hence wide bandwidth l!t.f/fo ,..., 1 at the price of reducing the bar's Q from
Q ,..., 105 to Q ,..., 2000 . The best amplifiers that could be impedance matched
to the piezo-electric transducers had rather large noise numbers. For these
first-generation bars the integrated cross-sections were J a0 df -::::::.
x
2 10- 2 1 cm 2 Hz, and the lowest detector noise temperatures were Tn -::::::. 4 K
corresponding to a minimum detectable burst amplitude with � year of
observation x
h 3 1yr "'3 10-16. Despite great effort in the early 70s by
excellent experimenters, there was great room for improvement. (For a
thorough revi e w of the first-generation experiments
1979; for other reviews, see Drever, 1977, Tyson and Giffard, 1978 and see Amaldi and Pi z ella,
Weber, 1986.)
In moving into the second generation, almost all the groups cooled their
bars to liquid helium temperatures (Tb = 1. 5--4 K rather than 300 K).
Gravitational radiation 40S
Several of the groups (M aryland , Stanford, LSU) constructed massive bars

(M - 2-5 tonnes) from a new alloy of aluminum that the Tokyo group
discovered has a Q 10 times higher than the old one (5 x 107 vs 5 x 106 at
104 K temperature ; Suzuki , Tsubono and Hirakawa , 1978). Perth and
Moscow, by contrast, chose to use exotic bar materials. Perth used a 1 . 5
tonne Niobium bar with Q = 2 x 108, while Moscow initially used a 1 0 kg
sapphire bar with Q 4 x 109 , and later when difficulty with cracking of the
=
sapphire occurred , Moscow switched to a 10 kg silicon crystal bar with

Q = 2 x 109 - silicon and sapphire because of their very large Q s , an
advantage bought at the price of low mass.
In the second generation each of the groups deemphasized piezo-electric
transducers and developed some variant of one or another of two new
transducer concepts : Moscow, LSU, Perth and Tokyo developed
parametric transducers in which the bar's vibrations with frequency
f0 - 10 3 Hz modulate the capacitance in a microwave cavity (or rf circuit),
thereby moving microwave photons from the frequency f0 - 109 Hz, at
which the cavity is driven, into side bands at f = f0 + f0 , at which the
amplifier measures the signal . The number of photons moved is
proportional to the amplitude of vibration of the bar. Stanford and
Maryland developed resonant transducers in which a mechanical
diaphragm , with a mechanical resonant frequency very nearly that of the
bar's fundamental mode, is attached to the bar's end. The vibration energy is
quickly transferred back and forth between the bar and the diaphragm, with
the diaphragm's displacement amplitude amplified over that of the bar by
jX diaph ragm I/I X bar l "' [(bar mass)/(diaphragm mass)]!. The vibrating
diaphragm modulates the inductance of a superconducting circuit, and a
SQUID (superconducting quantum interference device) amplifier is used to
monitor the current in the circuit. Rome developed a similar resonant
transducer with the diaphragm replaced by a toad stool which was one plate
of a capacitor and with a FET transducer used to read out oscillations of the
toad stool's voltage.
At present (late 1986) Stanford , Rome and LSU are all on the air with
systems that have resonant frequencies f0 "' 900 Hz, cross-sections J a df "'
0
(4 to 8) x 10 - 2 1 cm 2 Hz, and noise temperatures T0 "' (0.0 1-0.04) K.

Correspondingly the weakest burst they can detect with � year of
observation is h 3/y r "' 1 .0 x 10 - 1 7. Maryland and Perth are both likely to be
on the air with similar characteristics before this book is published ; and all
five groups are hoping to push their noise levels down to h 31yr "' 2 x 10- 1 8
within several years by further straightforward refinements. The
Guangzhou group, having only entered the field very recently, is still at
4 10 K . S. Thorne
room temperature, but with a sensitivity h 3;yr ':!:. 1 .6 x 10 - 1 6 , two times better
than that of any of the first-generation room-temperature bars (Hu et
al. 1986). The Moscow group with its small silicon and sapphire bars
operates at a much higher frequency than any of the other groups, fo ""
8 kHz; by the time this book is published they will likely be on the air with
h 3;yr "" 4 x 10 - 1 7 . The Tokyo group, on the other hand, has chosen a much
lower frequency than the others, f0 "" 60 Hz and is now operating a narrow
band search for periodic waves from the Crab pulsar (see below). These
sensitivities are shown in Fig. 9.4, along with those of other kinds of
detectors. For details of the present detector configurations and near-term
plans, see the gravitational-wave articles in the proceedings of recent
conferences (Ruffini, ed. , 1986; MacCallum, ed. , 1987).
(e) Sensitivies of bar detectors to periodic and stochastic waves

Although I have discussed the detectors' performances entirely in the
context of searching for short bursts of gravitational waves, bar detectors
can be used also to search for periodic gravitational waves and a stochastic
background at frequencies within the bandwidth !J.f of their sensors.
In a search for periodic gravitational waves, the experimenters will
typically use a bar with eigenfrequency fo slightly different from the
expected frequency f of the waves. The waves then will drive the
fundamental mode's complex amplitude X into oscillations with the beat
frequency f - f0. In searching for these oscillations the experimenters
integrate for a long time ; and correspondingly they can tum down the
coupling strength /3 of the transducer to the bar, and/or narrow the
bandwidth 1�.f, until the sensor's noise becomes negligible compared to the
thermal noise of the bar (see, e.g., Michel son and Taber, 198 1 , 1984). (The
maximum resulting bandwidth with the present Stanford detector would be
!J.f ':!:. 0.5 Hz). When this is done, the general formulas of Section 9.4.2(a) are
valid, with Sh(f) the spectral density of the bar's thermal noise, converted
into an equivalent gravity-wave spectral density
(89 )
and with the beam-factor averages having the values (79). In particular, the
[
detector's characteristic noise amplitude (equation (5 1)) is
G k Tb 1 1 f
c J ao df Q foi J '
h n = 15 3 (90)
and the brightest source that can be seen with 90 % confidence in i =! year of
( TKb ) (
integration is
h 3 /yr = 1 . 7h = 3 .9 x 10 - 25
n
1 )(
J ao df fo
)( )
1 10- 21 cm 2 Hz 1 1000 Hz i 107 1
Q
.
(9 1)
The Tokyo group is currently carrying out a search for gravitational
waves from the Crab pulsar using the above technique (Owa et al., 1986).
Their 74 kg , cryogenically cooled antenna has Q = 2. 1 x 107 , Tb = 4 K,
f0 = 60 Hz, and J a0 df � 2 .2 x l0 - 2 7 cm 2 Hz (so low because for
noncylindrical antennas with frequency fo lowered by special shaping,
J a0 df oc f5 ; Hirakawa et al., 1976). Correspondingly it has h 3 ;yr "'
3 x 1 0 - 2 2 . N o other present bars are optimized fo r periodic waves, since
there are no known sources in their frequency bands ( - 900 Hz and - 8 kHz).
However, with the technology of present burst-optimized bars it should be
possible to achieve the thermal-noise-limited sensitivity (9 1) with Tb = 4 K,
J a0 df� 8 x 10 - 2 1 cm 2 Hz, f0 � 900 Hz, and Q � 5 x 106 corresponding to
h 3 ;yr � 4 x 1 0 - 2 5 (Stanford, LSU, Rome, Maryland); and Tb = 4 K,
J a0df � 2 x 1 0 - 2 3 cm 2 Hz, f0 � 8 kHz, and Q � 2 x 109 corresponding to
h 3 ;yr � 1 .4 x 1 0 - 2 5 (Moscow). See Fig. 9.6.
When searching for stochastic background it is desirable to open up the
bandwidth �.f until the sensor noise becomes almost as large as the bar's
thermal noise (�.f � 0.5 Hz for the present Stanford bar). With Sh (f) then
given (approximately) by the thermal-noise spectral density (equation (89)),
the noise amplitude hn and the 90 %-conficence !-year sensitivity h 3 ; yr for
[ 1
stochastic background become (equations (66) and (67))
G k Tb 1 1
h n (f) � 1 5 C 3 (92)
J a0 df Q J(tf df) '
h , ,, , = l .7h. - s x 10- 2V�Y( ;�:�; Y (1� )t��zy.
2
10 - Hz 1
(93)
This h 3 ;yr is shown in Fig. 9.7 for the parameters of 1987 bar technology
(those given in the last paragraph, plus llf � 1 Hz). For details of searches
for stochastic background that were carried out using first-generation bar
detectors, see Hough et al. (1975) and Hirakawa and Narihara (1975). For
discussions of sensitivity that are more detailed and sophisticated than the
above sketch, see Hirakawa, Owa and Iso ( 1985), Weiss ( 1979) and
references therein.
(f) Q uantum limit and quantum non-demolition

Quantum mechanics constrains the sensitivity that can be achieved by bar
4 12 K . S. Thorne
detectors using the present kinds of sensors : the fundamental mode of a bar,
being highly decoupled from the rest of the world, can be regarded as a
simple harmonic oscillator with mass M eff and angular frequency w 0 = 2rrf0 .
As such , it is subject to the laws of quantum mechanics for oscillators : its
generalized position x and momentum p must be regarded as hermitian
operators that fail to commute, [x , p] = i h ; and, correspondingly, the real
and imaginary parts of its complex amplitude,
X 1 = x cos w 0 t - ( ) p
Merrill o
sin w 0 t,
X 2 = x sin w 0 t +
Merrill o
( )
cos w 0 t ,
p
(94)
are non-commuting hermitian operators with commutator
(95)
From this commutation relation we infer, via the Heisenberg uncertainty

principle, an absolute limit on the variances of X 1 and X 2 in any quantum
mechanical state (Thorne e t al. , 1978) :
h
1'1 X i 1'1 X i � 2 (96)
M errWo
The principles of quantum mechanics guarantee that one can never measure
X 1 and X 2 simultaneously with a precision that violates this uncertainty
principle. In fact, it turns out (Caves, 1982 ; Yamamoto and Haus, 1986) that
even the most ideal of measuring systems will introduce additional noise
equal to (96), thereby producing a lower bound
(97)
on the products of the rms noise in the measured values of X 1 and X 2 .

The sensors used in the present generation o f bar detectors measure X 1
and X 2 simultaneously with equal accuracies ; and, correspondingly, in
searches for short bursts they are subject to the 'standard quantum limit'
(Braginsky and Vorontsov, 1974 ; Giffard , 1976)
Tn = "!M err w 5 [(1'1 X 1 ) 2 + (1'1 X 2 ) 2]/k � hw0/k = 4.8 x 10 - s K ( fo

1000 Hz
) .
(98)
Notice that this standard quantum limit places the severe constraint
h 3 /yr "" (
> 4.4 x 10 - 2 0 )(
fo
1000 Hz J a df
0
)
1 10 - 2 1 cm 2 Hz 1
(99)
on the detector's burst sensitivity (82). It is fairly likely, though far from
certain, that the strongest kilohertz-frequency bursts striking the earth three
times per year have characteristic amplitudes h e < 10 - 2 0 (see Section 9.4. 1);
and , correspondingly, it may turn out to be crucial for bar detectors of the
future to circumvent the standard quantum limit (98) , (99).
The uncertainty principle (96) suggests a promising method for
circumventing the standard quantum limit (Thorne et al. , 1978 ; Braginsky,
Vorontsov and Khalili, 1 978) : one should devise a new kind of sensor that
measures X 1 with high accuracy, while giving up accuracy on X 2 . Such
sensors, called 'back-action-evading sensors' (a special case of 'quantum
non-demolition sensors'), are now under development in a number of
laboratories (Braginsky, 198 3 ; Bocko and Johnson , 1 984 ; Oelfke, 1983 ;
Blair, 1 982) ; and they may make possible bar sensitivities in the 1990s that
will beat the standard quantum limit by modest factors. For a detailed
review of quantum non-demolition measurements - i.e. measurements that
do not change the quantum state of the system being measured - see Caves
( 1 983).
Although a back-action-evading sensor gives up accuracy on one of the
wave's two quadrature components, that accuracy can be regained by
looking at the same wave with two different detectors : on one detector, with
complex amplitude X = X 1 + i X 2 , measure X 1 with high accuracy and X 2
with poor; on the other, with complex amplitude Y = Y 1 + iY2 , measure Y2
with high accuracy and Y1 with poor. From X 1 infer the detailed evolution
of one of the wave's two quadrature components; from Y2 infer the
evolution of the other. In this way , in principle, the quantum mechanical
properties of the detector can be completely circumvented and the only
constraints of principle on the accuracy of measurement are associated with
quantization of the waves themselves. For further discussion and details see
the reviews by Caves et al. ( 1 980) ; Caves ( 1 983) ; Braginsky, Vorontsov and
Thorne ( 1 980) .
It is worth noting that a back-action-evasion measurement, ideally
performed, should drive the bar's fundamental mode into a 'squeezed state'
(Hollenhorst, 1 979). Squeezed states have been studied extensively in recent
years in the context of quantum optics (see, e.g. Schumaker, 1986 ; and
Walls, 1 983) ; and we shall return to them in Section 9.5.3(f) below when
discussing beam detectors.
4 14 K . S. Thorne
(g) Looking toward the future

It may be that bar detectors in the distant future will find their greatest
applications at much lower frequencies than are now common : at lower
frequencies the bar can be much longer and more massive, and
correspondingly more sensitive; and good bandwidth, !J.f "' f0 will
correspond to a longer averaging time and thus will be easier to achieve. For
discussion, see Michelson ( 1986).
Although individual bar detectors in the foreseeable future will be limited
to moderately small bandwidths !J.f/f ;S 0.2, and therefore will be able to
acquire only very limited information about the wave form hJkT(t) of a
gravity-wave burst, once waves are being detected in profusion it may be
possible cheaply to replicate the detectors with a variety of sizes and hence a
variety of fundamental-mode frequencies , thereby creating a 'xylophone' of
networked detectors with a large overall bandwidth (Michelson and Taber,
1984).
9 .5 .3 Beam detectors
(a) A brief history of beam-detector research
The germ of the idea of a laser-interferometer gravitational-wave detector
('beam detector') can be found in Pirani ( 19 56) ; but - so far as I am aware -
the first explicit suggestion of such a detector was made by Gertsenshtein
and Pustovoit ( 1 962). In the mid- 1960s Joseph Weber, unaware of the
Gertsenshtein-Pustovoit work , reinvented the idea but left it lying in his
laboratory notebook unpublished and unpursued. In 1970 Rainer Weiss at
MIT, unaware of Gertsenshtein-Pustovoit or Weber, reinvented the idea
and carried out a detailed design and feasibility study (Weiss, 1972) in which
many of the techniques now being used were conceived . Unfortunately,
Weiss was unable to obtain funding to push forward with a significant
experimental effort.
Robert Forward at Hughes Research Laboratories in Malibu , California,
having learned the concept of the beam detector from Weber (his former
thesis advisor), was motivated indirectly by Weiss in 197 1 to construct a
prototype detector with funding from Hughes. By 1972 Forward and his
colleagues at Hughes were operating the world's first prototype beam
detector - an instrument that demonstrated the idea could really work, and
that was remarkably sensitive considering the modest effort put into it :
[Sh (f )J � � 2 x 10 - 1 6 Hz- � between 2500 Hz and 25 000 Hz, corresponding
to h 3; y � 1 x 10 - 1 3 for 2500 Hz bursts (Moss, Miller and Forward, 197 1 ;
r
Forward and Moss, 1 972 ; Forward, 1978) . Regrettably, Forward could not
obtain funds to move from this first prototype to a more sophisticated
instrument; so his project was shut down.
With the completion of the first generation of bar detectors in 1975, each
experimental group that decided to stay in the field looked carefully at a
variety of possibilities for sensitivity improvement. While most groups
decided to stick with bars, two switched to beam detectors : Munich, led by
H . Billing, and Glasgow , led by Ronald Drever with Jim Hough second in
command. The Munich group was strongly influenced by a proposal to
develop beam detectors that Weiss had submitted to NSF , and that NSF
had refused to fund ; and so Munich pushed forward (Winkler, 1977) along
the lines that Weiss had hoped to follow , using a Michelson interferometer
design (see below) . The Glasgow group first built a small Michelson
interferometer (Drever et al. , 1 977), then switched in 1 977 to a new Fabry
Perot design invented by Drever (Drever et al., 1980) .
In 1 979 Caltech managed to attract Drever away from Glasgow (part
time at first, full-time later) , leaving Hough as the Glasgow leader. At
Caltech Drever started up a beam-detector project ; and NSF, finally
recognizing that beam detectors were worth funding, agreed to support both
Weiss at MIT to develop his original idea of a Michelson system and Drever
at Caltech to develop his Fabry-Perot system . More recently, in 1 983 , Alain
Brill et initiated a beam-detector effort in Orsay, France (Brillet and
Tourrenc , 1983; Brillet, 1 985).
Munich, Glasgow , Caltech and MIT all now have working beam
detectors with amplitude sensitivities "" 2000 times better than that of
Forward's first prototype but "" 5 times worse than the best bars. These
detectors are small-scale ( 1-40 m) prototypes for the full-scale (several
kil ometer) beam detectors that will be required for real success. Design and
costing studies are now underway for the full-scale systems (called 'Laser
Interferometer Gravity Wave Observatories' or LIGOs) ; for details of these
studies, see Linsay et al. ( 1983), Drever et al. ( 198 5), Maischberger et al.
( 1985), Winkler et al. ( 1 986), Hough et al. ( 1986) . There is hope that full
scale LIGOs will be constructed in the late 1 980s and early 1 990s and will be
operating in the mid- to late- 1 990s with sensitivities in the region where
gravity waves are expected.
(b) How a beam detector works

The current and planned beam detectors are designed to operate at
frequencies below 10 kHz because astrophysical arguments suggest that the
4 16 K. S. Thorne
waves will be weak above this frequency (see page 158 of Thorne, 1978) ; and
they have their best sens1tiv1t1es at frequencies below 1 kHz.
Correspondingly, the waves they seek all have reduced · wavelengths
X > 5 km and most have X > 50 km. Since the planned detectors all have sizes
L � 4 km, the condition L � ): for use of a 'proper-reference-frame analysis' is
satisfied, though only marginally in extreme cases. I shall use such an
analysis in the discussion below. For an outline of the alternative, TT
analysis, see, e.g., Exercise 37.6 of MTW.
A beam detector consists of one or more receivers that are operated
simultaneously, with cross-correlated outputs - the cross-correlation , as
usual, being the key to removing spurious, non-Gaussian noise. A simple
version of a Michelson-type receiver is shown in Fig. 9.9, three
dimensionally in part (a) and as seen from above in part (b). (Ignore for the
moment the propagation and polarization pieces of part (a ) . ) The receiver
consists of three masses which hang on wires from overhead supports and
swing like pendula. The masses are arranged at the ends and corner of a
right-angle L. When a gravitational wave propagates vertically through the
receiver with polarization axes along the L ( + ' polarization), its '
Fig. 9 . 9 . Schematic diagram o f a Michelson-type beam receiver for

gravitational waves (part (b)), and of the waves' propagation and
polarization angles (0, </J, t/I) relative to the receiver (part (a) ; cf. Fig . 9.2).
END
MASS
END
MAS S
RECYCL I NG
M l R ROR /�( __ _ _ _
�rifv1fIT�=�:;::==:::::=l >-'.<:
, :
LASER
'
�-�
BEAM SPL I TTER --- L ----
(b) PH OTO D I O D E
quadrupolar force field (5) pushes together the masses on one arm of the L
while pushing apart the masses on the other arm . In the next half-cycle of the
wave, the directions of the pushes are reversed. Since the waves being sought
have frequencies f far above the 1 Hz swinging frequency of the pendula, the
pendular restoring forces have no opportunity to make themselves felt : the
masses respond to the gravity-wave pushes as though they were free. With
the origin of the p roper reference frame placed on the central mass, the
central mass is left unaffected while the end masses oscillate longitudinally
with displacements
c5x(t) = 1Lh + (t) for mass on x axis, ( lOOa)
c5y(t) = -1Lh + (t) for mass on y axis ( lOOb)
(equation (6)). Here L is the (approximately equal) length of each arm.
Correspondingly, there is an oscillation in the difference l(t) of the arm
lengths, Dl(t) = Dx(t) - c5 y(t), given by
c5l ( t) = h + ( t)L . ( 10 1)
It is straightforward to show that in the more general case of a wave which
impinges from a direction ( 8, </>) on the sky with polarization axes rotated at
an angle t/I relative to the constant-</> plane (Fig. 9.9(a)), the difference in arm
lengths l oscillates as
c5l(t) = h(t)L, ( 102)
where h(t) has the standard form (26)
h(t) = F + ( 8, </> , t/l )h + (t; z, /3) + F ( 8, </> , t/l )h (t; z, /3),
x x
( 103)
with beam-pattern factors (cf. Forward, 1978 ; Rudenko and Sazhin, 1980;
Estabrook, 198 5 ; Schutz and Tinto, 1987)
F + (8, </>, t/I) = 1( 1 + cos 2 8) cos 2</> cos 21/1 - cos 8 sin 2</> sin 21/1,
( 104a)
F ( 8, </>, t/I) = 1( 1 + cos 2 8) cos 2</> sin 21/1 + cos 8 sin 2¢ cos 21/1.
x
( 104b)
The difference of arm lengths l(t) is monitored by Michelson
interferometry : a beam splitter and two mirrors are attached to the comer
mass as shown in Fig. 9.9(b), and one mirror is attached to each end mass. A
laser beam shines through a hole in the comer ma'ss and onto the beam
splitter, which directs half the beam toward each end mass. The mirrors on
the end masses reflect the beams back toward the comer-mass mirrors,
which in turn reflect the beams back to the end masses, which reflect the
beams back through holes in the corner-mass mirrors and onto the beam
4 18 K . S. Thorne
splitter where they are recombined. Part of the recombined beam goes out
one side of the beam splitter toward the laser (ignore for now the 'recycling
mirror' in (b), it is absent in the simple version of the receiver being discussed
here); the other part of the recombined beam goes out the other side toward
a photodetector. Oscillations in the arm-length difference '5l(t) produce
oscillations in the relative phases of the recombining light, and thence
oscillations in the fraction of the light which goes to the photodetector
versus that which goes back toward the laser. The photodetector, by
monitoring the oscillations in received intensity, in effect is monitoring the
.
oscillations '5l(t) of arm-length difference and thence the gravity-wave
oscillations h ( t)
In practice the laser beams are made to bounce back and forth in the arms
not just twice as shown in Fig. 9.9(b), but rather a large number of round
trip times B, making B distinct spots on each end mirror. In the simple case
that the gravity-wave-induced arm-length difference '51 = Lh does not change
much during these many round trips (see subsection (e) below for the case of
large change), the bouncing light beam will build up during its B trips a total
phase delay
( 10 5 )
where ,te = A e /2rr is the reduced wavelength of the light (,te = 0 . 08 18 microns
for the light from the argon ion lasers currently being used). This phase delay
can be monitored, by the photodetector, with a precision i1 <1> = l/(N/7)!,
where N Y is the total number of photons that the laser puts out during the
time f over which the photodetector intensity is averaged, and Y/ is the
photon counting efficiency of the photodetector (Y/ - 0 4-0 9 ) . When
searching for a gravity-wave burst with characteristic frequency f, it is
. .
optimal to average the photodetector intensity for half a gravity-wave
period, f - 1/2f; and correspondingly the phase delay can be inferred with a
photon-counting-no ise ('shot-noise') precision
L1<1>
sho t=
1
(Ny Y/ )!
(-
-
he/,te ) !
J Y/( 1/2J) ' ( 106)
o
where /0 is the laser output power. By comparing equations ( 105 ) and ( 106)
we obtain a rough estimate of the amplitude of a gravity-wave burst that
produces a signal of the same strength as the rms shot noise
h ,..., [2hcXe (2BL) 2Jt

sho t =
I 011
f
,..., 7.2 x 10_ 2 1

km ( ,te )t( IO Watts)t( f )t · ( l07)
50 1
B L 0.082 µm I 017 1000 Hz
This formula and these
sensitivities of beam receivers. numbers give some indication of the potential
Fabry-Perot
sensitivities as Mi beam
c helson recei v ers
receivers. have
Fig. essentially the same
9. 10 shows a Fabry-Perot receiver
potential
vibroken
ewed from above. Here, by contrast wi t h Fig. 9 .9, the corner mass has been
into three separate pieces, one carrying each mirror and one carrying
thereduces
beamspurious
splitter.forces
This breaking
on the up of
mirrors; the
si n comer
ce mass, initiated in
1985 it has been standard in all
Muni c h,
beam receivers. In the Fabry-Perot system of Fi g . 9. 10 each arm is operated
assplitter
a resonant
and Fabry-Perot
enters the two cavity:
cavities light
through fromthethe laser
backs of i s spl
the i t
comer at the beam
masses'
partially transmitting mirrors. The two arms are
equilibrium lengths very nearly equal to a half-integral multiple of the arranged to have
wavelength 2nXe of the laser light. Consequently, the entering light finds
itself in resonance with a mode of each 'cavity; and it gets resonantly
Fig . 9. 10. Schematic diagram of a Fabry-Perot-type beam receiver for
gravitational waves.
COR N E R END
R ECYC L I N G MASS MASS
M I RROR
.: • : ·
· · · . '. ·
.
Re t'RE
PHOTOD I O D E
420 K . S. Thorne
trapped
toward in
the the
beam cavi t y,
splitter. building up to high intensity before exiting back
Slight
resonance changes
and in
thereby the length
produce of each
sharp cavity
changes drive
in thethe cavity
phase of slightly
the off
exiting
light. Consequently, when the exiting light beams from the two cavities
recombine at the beam splitter, their relative
slight modulations '51 of the two cavities' length difference l; and phase is hi g hly sensitive to
correspondingly 1
the intensity of light onto the photodetector is highly
sensitive to '5 . If the comer mirrors have a probability for reflecting photons
f7tc and a transmission probability 1 - f7tc (and no scattering), and if the end
mirrors
descri b edreflect
by the much
same more
formulaseffici e ntly, then this Fabry-Perot sensitivity
( 105)-( 107) as for a Michelson receiver, wit h
is
the number of round-trips B in each Michelson arm replaced by 4/( 1 - f7tc) :
--+
B 4/(1 - f7tc). ( 108)
Itreceiver
is also possible
in an (and,
alternativein fact, is
mode current
where,practi c e)
instead to operate
of a Fabry-Perot
recombining and
interfering
one arm the
and beams,
the dif f the
erence laser' s frequency
between the is locked
laser' s to the
frequency eigenfrequency
and that of of
the
other arm is the gravity-wave signal; see, e.g. Hough et al. ( 1983) or Spero
( 1986a) for details. This mode of operation is technically easier than beam
recombination, but the ease is bought at the price of some debil itation in the
ultimately achievable shot noise.
(c) Noise in beam detectors
Photon shot noise i s but one of many noise sources
detectors. Almost always, thus far, the other noise sources are so strong that that plague beam
the effects of shot noise are lost amidst them. Typically the experimenters
struggle
shows up;for a
thenlong time
they to reduce
improve thethe other
shot noises
noise suffi
somewhat c iently
by that shot
increasing noise
the
laser power; then they begin a long struggle once
sources. In this way the overall gravity-wave amplitude noise at kilohertz again with the other noise
frequencies has been reduced during the peri o d 1980-86 by a factor 1000 ,...,,
(106-fold improvement in energy noise).

Among the most serious of the other noise sources are the foll owing:
amplitude and phase fluctuations in the laser output; imperfect matching of
the laser beam into the Fabry-Perot cavities or the Michelson mirror system
(iimmperfect
perfections in beam directi o n, beam shape, and
matching of the wave fronts of the recombining beams at the beam beam wavelength);
splitter; fluctuations in the index
(even though the arms are inside vacuum pipes, residual gas can cause of refraction of the gas i n side the arms
problems); scattering of light from one part of the optical system into
another; thermal noise (thermal vibrations)
and in the wires that suspend them; imperfect alignment of the mirrors; and i n the end and corner masses
seismic
the wires and fromacoustic which noise
the from
masses the
hang.outsideAlmost world,al l which
of cause
these eff e vibrations
cts produce ofa
displacement noise c5l that is independent of the arm length L; and,
correspondingly, their effects on gravity-wave amplitude sensitivities scale
as h oc 1/L. It is this scaling that motivates the move from short prototypes
to long LIGOs.
A
light beam
into recei
the v er
photodetector is intrinsically broad band: its output, the intensity
Ipd (t), is averaged (by filtering) over the shortest
of the
timescale, f,... , ., 10 - 4 s, that gravity waves are likely to contain, and then is
recorded
data for for
si g future
nals of analysis.
al l The
frequenci future
e s f analysis can include searchi n g i n the
� 1/f,....,, 104 Hz, and for signals with
complicated time dependences that
Correspondingly, a beam receiver's net noise depends on the frequency fatembody a broad range of frequenci e s.
which one studies the recorded output.
which More are speci
removed fi cally, by' thecoi noise
n ci d ent (excludi
operati n og n non-Gaussian,
of two spurious
independent recei vevents
ers) i s
characteri
dih(t)fference z ed by the spectral densi t y S1 ( f) of the output-inferred arml e ngth
l; or, equally well, by the spectral density Sh(f) of the gravity wave
(equation (103)) . Because the effect of the gravity wave on the arm
length diff e rence is c5l(t) = Lh(t), Sh(f) is also called the spectral density of
strain, and these two spectral densities are related by
(109)
Itgenerally
is Sh(f) quote or its square root, or S1 ( f) or i t s square
when discussing the overall performance of their beam root, that experi m enters
detectors. 4
In Sections 9. 4 . l(b ), 9. 4 . 2 (a) and 9.
of Sh(f) for the minimum-amplitude signal h 3/yr that can be detected with . 3 (a) we deri v ed expressions i n terms
90 % confidence in a � year search using a detector composed of two
identical cross-correlated receivers. Those expressions involve averages over
the receiver' s beam-pattern factors. For
9. 9 and 9.10, with beam-pattern factors (104), the appropriate averages are the L-shaped beam recei v ers of Figs.
(F� ) = (F2,, ) = i, (110)
422 K . S . Thorne
and, correspondingly ,
h31yr ( fc) = 1 l [fcSh ( fc)]� for bursts (equation (34)) , ( 111)
h3/yr (f) = 3 . 8 [Sh (f) x 10 - 7 Hz] l
for periodic waves (equation (52a)), (112)
( llf )-� [fSh(f)]
h3/yr (f) = 4 . 5 1 0 - 7 Hz
1
i
for stochastic waves (equation (67)). ( 1 1 3)

Here in ·the burst equation the numerical factor in equation (34) has been
taken to be 5 corresponding to a characteristic frequency fc ,..., 10 Hz to
104 Hz - the largest band that earth-based receivers are likely ever to
operate in; and
phase1)-(are1 13)known in the
in periodic
advance equation
(equati o n it
(52a)is assumed
rather that
than the frequency
(52b)). Expressions and
(11 are the bases of the beam-detector noise amplitudes shown in
FigWhens. 9.4,the9 .6twoandrecei
9.7.
vers that make up a detector do not lie in the same
plane
arms (because
parallel, of curvature
their joint of the earth
sensitivity between
is them)
somewhat or do not
debilitated. have their
Detailed
analyses of this have been carried out by Whitcomb and Saulson ( 1984) and
by Schutz and Tinto ( 1986).
(d) The noise in the present prototypes
At present the MIT1.5 m;groupthe Muni is developing Michelson recei v ers using a
prototype with L= ch group is developing Michelsons using
L = 30 m (Shoemaker et al., 1986); the Glasgow group is developing Fabry
Perot receivers using L = 10 m; and the Caltech group is developing Fabry
Perots using L = 40 m. Although the arm lengths and optics of these four
prototypes are very different, their displacement sensitivities are roughly the
same-and have been so throughout the past six years, during which all have
improved roughly in step by about three orders of magnitude. Throughout
these improvements Munich has maintained a slight (factor 2 or 3 in
amplitude) lead over the other groups. (Note added in proof. Since this was
written Glasgow has forged ahead of Munich by a factor 3 in displacement
sensitivity, making them equal in h sensitivity.) Fig. 9.11 shows the spectral
density of strain noise for the Munich prototype as of February 1986
(Schilling, 1986). ' The noise is due, predominantly, to seismic noise
(iandnadequate i solation) below 250 Hz, probably seismic
1000 Hz, photon shot noise between 1000 and 6000 Hz, and thermal
noise between 250
noise in the mirrors and pockels cells above 6000 Hz. In Fig s. 9.4, 9.6 and 9.7
(upper right) are shown the sensitivities h 31yc (equations ( 1 1 1)-( 1 13)) for the
Munich and Caltech prototypes in 1986 (Schilling, 1986 ; Spero, 1986b) and
Glasgow in early 1987. As an illustration of the sensitivity progress, Fig . 9.4
also shows h 3;yr for bursts in the Munich and Caltech prototypes a few
months after each was first turned on ( 1980 and 1983).
(e) Spectral density of shot noise for simple, recycling and resonating
receivers
In the present prototypes it is advantageous to store the light in the arms as
long
gravi tas possible,
y-wave thereby
sensitivi t y; cf.building
the up
B-dependence the largest
in possible
equations phase shift and
( 105) and ( 107).
However, in a kilometer-scale LIGO one easily can store the light longer
than half a gravi t y-wave period, i. e . for more than B = 75(1 km/L) x
( 1000 Hz/f) round-trip traverses of an arm. Such long storage is self
defeati n g: the phase shi ft built up so laboriously duri n g
ofsigthen. wave gets removed during the second half-period because h(t) reverses the first half-peri o d
This shows up clearly i n the spectral densities of shot noise Sh(f) for the
idealized Mi c helson and Fabry-Perot receivers of Fi g s. 9.9 and 9. 10 (stil
without the 'recycling mirrors' in place). Assuming as above that the
Michelson
and the mirrors
Fabry-Perot have
end negligible
mirrors losses
have during
negligible B round
transmissiontrips in each
compared arm,to
the corner mirrors
1 - � E � 1 - �c = 4/B ( 1 14)
(cf. equation ( 108)), the spectral densities of shot noise are (cf. Giirsel et al.,
Fig. 9 . 1 1 . Square root o f spectral density o f noise [Sh(J)]1 plotted against
frequency f for the Munich Michelson-type beam detector with 30 m arms,
as of February 1986.
IB
3 X l0 -
C\J
10- I B
'
I
' N
3 x10- 19
� 10- 19
3 x 1 0 - 20
0 500 1000 2000 3 0 00 4 000 5 0 00
f, Hz
42 4 K. S. Thorne
1983; Meers, 1983; Brillet and Meers, 1987)

Sh = 2 hc J:. --
e( 1 2 ) 1 (2nBLf/c)+ 2 for Fabry-Pero� (115a)
(f)
1017 2BL
x
[sin(2nBLf
2nBLf/c ]2 for Michelson. (115b)
/c)
These spectral densities are shown in Fig. 9.12 as a function
(light storage time)/(gravity-wave perfod), with (gravity-wave period) = 1/f of 2BLf/c=
helshotd fixed. When (storage time)/(period) 1 the two� receivers have the same
noise, and that shot-2 noise improves with increasing storage time as
Sh (f) oc (storage time) • When (storage time)/(period) � 1 the Fabry
Perot shot noise stops
aNoteseriesthatof theoscillationsimproving,
with minimawhile the
equal Michelson
to the shot noise
Fabry-Perot undergoes
shot noise.
minima of the Mi c helson oscillations occur when 2BLf
(light-storage time)/(gravity-wave period) = 0. 5 , 1.5, 2. 5 , . . . , while their /c =
Fig . 9. 12. The improvement of photon shot noise with increasing number of
bounces of light in a simple beam receiver. The receiver is of the Fabry-Perot
(Fig. 9. 10) or Michelson (Fig. 9.9) type with recycling mirrors absent, with
1 -�E � 1 - �c = 4/B in the Fabry-Perot case and with B round trips in each
arm in the Michelson case. Plotted vertically is spectral density of shot noise
in units proportional to f 2 but independent of the light-storage time 2BL/c.
Plotted horizontally is (light-storage time)/(gravity-wave period) = (2BL/c)f.
The Fabry-Perot shot noise is given by the solid curve (equation ( 1 15a)); the
Michelson by the dashed curve (equation ( 1 15b)).
I I I I
I
I
I I I I
I
I I I I I
4 I I
I I I
I I I I I
C\J 21 I I I
'
- � lu
-
-
�
...--....
3 gl
�I
QI
I
I
I I
'
I
I
I
I
I
I
I
Cf) .K
� Q)
� I I
(.) _o
�I I I
.c
2 I I \ I
I
C\J I \ \
I I
I \ I \ I
' / \ I \ I
FA BR Y- PE
RO T
_..
0
0 2 3 2 8 Lf
c
maxima (Sh = oc ) occur when the lig ht is stored for an in tegral number of
gravity-wave periods. This i s just what we would
h(t) reverses sign every half-period, thereby removing during a second half
expect from the fact that
period the signal put onto the li g ht duri n g a first
Perot receiver there are no oscillations of Sh(f) because different photons hal f -peri o d . In the Fabry
experi e nce di f
the reflectivity f!ltc.ferent storage times-i. e . because of the probabi l istic nature of
Drever (1983) has devised a method for improving the sensitivity of either
aforMimuch
chelsonlonger or athan
Fabry-Perot when mirror reflecti v iti e
a half-period. The basic idea is to extract the light afters permit stori n g light
athehalf-peri
cavity o d,
along when
with further
and in storage
phase is
withsel f-defeating,
new l a ser and
light. then
More rei
speci n sert
fi i t
cally, i n to
for
the Mi c helson of Fig. 9.9(b), one adjusts the relative arm lengths so that very
little of the recombined light emerges from the beam splitter toward the
photodiode
operation thatwhere the
optimizes gravity-wave
the sensitivity, si g
itnal i
turnss read
out). out
Then (a
most mode of of
the
recombined
inserted to lig
direct ht emerges
that emergent toward l i ghtthe
back laser;
into and
the '
beamr ecycl i ng
splitter mirrors'
along wi are
t h
fresh laser
withAs aansingle light. For
recyclingthe Fabry-Perot
mirror. of Fi g . 9.10 the same effect is achi e ved
aid i n quanti f yi p g the recycli n g-i n duced
consider the (realistic) situation in which the technology of mirror(0.9999 i m provement in shot noise,
coatingsat
limits the mirror reflectivities to some maximum value f!ltmax
present; perhaps 0.999 99 a few years from now). It obviously is optimal to
place at the end of each arm a mirror with this maximum, so f!ltE = f!ltmax ·
Consider, for concreteness, the Fabry-Perot recei v er of Fi g . 9 . 10. If the
corner mirrors are chosen also to have the maximum reflectivity, f!ltc = f!ltfa
then
gained essentially al l the unused lig ht leaks out the
by recycling. In this case of a simple, non-recycling Fabry-Perot withend mirrors and nothing i s
=
.1?-?c fAE = «1'm ax the spectral density of shot noise is (cf. Giirsel et al., 1983;
Meers, 1983 ; Brillet and Meers, 1987)
( l 16a)
where
2hcJ:.e (1-f!lt E)2
So =
= [ 3 . 6 10 - 2 5 ( ) 1( 100 W)-i(l -f!lt E) (l km)] 2

1011 2L
x J:.e
(Hz)1 0.08 1 8 µm 1017 10 - 4 L ' (1 16b)
426 K. S. Thorne
fo -
_ (1 -�E)c 2 .4 Hz( 1-�E
4nL 10-4 L ) ( l km) . ( 116c)
(Note
worse that,
than because
the � �� so much
limit lig
ofht i s
equationlost out the
(115a) end
- mirrors;
worse by athi s
factor noise
4 at
is
frequenci e s c E
f � f0 and by a factor 16 at f �f0 .) This non-recycled (' simple' )
Fabry-Perot
An experi mnoise
enter is
who shown
wishes as a
to solid
improve curve on in Fig.
this 9
noise. 13.level by recycl i n g
must choose a frequency fk �f0 near which the noise level is to be
minimized. The minimal noise level near f=fk will then be achieved by
setting (117a)
so(nfk)the- 1 effective storage time in each arm is l
2BL/c = 8Lc - /( 1 - �d =
= x
(l/n) (period of a gravity wave with the optimal frequency fk).
Mini m al noise also requi r es a speci a l choice for the reflectivity ,gfR of the
recycling mirror (Fig. 9.10)
1 - �R= 4(1(1--��E)c) . (117b)
With these choices of reflectivity, the light-recycling Fabry-Perot receiver of
Fig. 9.10 has shot noise (cf. Giirsel et al., 1983; Meers, 1983; Brillet and
Meers, 1987)
s,(f) = � s. [ 1 + (jJJ (117c)
.
This noise is depicted in Fig. 9.13 for two choices of fk (dashed curves). Note
that this noise has a ' k nee' at the frequency fk; for this reason fk is called the
knee frequency. Note further that recycling produces an overall
inon-recycl
mprovemented inFabry-Perot
Sh(f), at frequenci e s f � fk, by a factor f0 /2fk relative to a
with equal reflectivities for corner and end
mirrors (equations (116)) and an improvement by 2f0 /fk relative to the very
best non-recycl e d Fabry-Perot - one wi t h 1 -�max-:- - 1 - �E � 1 - � c �
8nLfk/c (equation (l 15a)). This improvement at f � fk is bought at the price
ofoverworsened noise at f� (fkf0 /2)1 . Note further that the improvement factor,
an optimal non-recycle� dcycFabry-Perot at f � fk, is
S e led 2fo 1 - �E
�o n-r cyc
s e led fk 1 - �c '
= (118)
-
whi c h is 1/(the
lostcbyal intuition
isphysi mean
leakage through numberthe of times
high-reflecti thatvthe
i t y light
end can be
mirrors). recycl
This e
isd before
just what it
should suggest.
For the recycl e d Michelson receiver of Fig . 9. 9 , recycling produces

essentially the same Sh ( f) as for the Fabry-Perot of Fig. 9.10 at frequencies
f ;::; fk · However, above the knee frequency the recycled Michelson exhibits a
Fig. 9 . 1 3 . Spectral density of shot noise fo r Fabry-Perot beam receivers

operated in various optical configurations. The vertical and horizontal scales
are both logarithmic. The solid curve is for a simple Fabry-Perot with all
mirrors - corner and end - having the maximum achievable reflectivity, 9fc =
£1,fE £1,f max (equations ( 1 16)). The dashed curves are for a Fabry-Perot with
=
light recycling (equation ( 1 17c)), in which the end mirrors have the maximum
achievable reflectivity £1,fE = £1,fmax • the corner mirrors are adjusted to produce
the desired knee frequency (fk i and fk z for the two curves shown ; equation
( 1 17a)) , and the recycling mirror is adjusted to minimize the noise at the knee
frequency (equation ( 1 17b)). The dotted curves are for a Fabry-Perot with
light resonating (equation ( 1 19c) ; Fig. 9. 14) in which the end mirrors have t he
maximum achievable reflectivity £1,fE £1,fmax ; the corner mirrors are adjusted
=
to produce the desired resonant frequency for gravity waves (fR1 and fRz for
the two curves shown ; equation ( l 19a)); and the recycling mirror is adjusted
to minimize the noise at the resonant frequency (equation ( 1 19b)).
/
/
/
/
/
. . /
/
;� /
�
�I.Po� /\[ //
/ : /
Y'.)o. : /
� :
/ : /
R E CYCLING ·�
- - - - - - - - - ·.- - ·:- - - -/- -
. / --(
: / :
.
: / :
·. ; / :
. . .
.
�. .: ..,..... /
RECYC L I N G
- - - -- _....,. :. : .
.....fi--.6f = f0
0. 5 fk l fk l 2fkl 0. 5 fk 2 fk 2 2 fk 2 In f
fRI fR 2
428 K . S . Thorne
sequence of noise peaks and troughs analogous to those for a non-recycled

MichelOf al s
l on
opti(equation
c al (115b)
configurations and
yet Fi g . 9.12).
invented, recycling i s the best when one is
searchi n g for broad-band gravitational waves (waves with llf ;;:, f).
However, when one is searchi n g for narrow-band waves (waves with /J.f 4:. f,
e.configurati
g. the periodi c waves from a pulsar), recycl i ng
on called light resonating (invented by Drever, 1983), which isgives worse noise than a
shown
Michelson i n Fi g
case . 9.14.
but i sAs usual,
descri b ed the
by me�hod
the simplest is conceptually
formulas i n simplest
the in
Fabry-Perot the
case.The basic idea, as shown for a Michelson in Fig . 9.14(a), is to store the
l(i.igeht. forin anB =armc/4fforLoneround half-cycl e of the gravity-wave frequency of interest fR
R trips); then, instead of extracti n g i t through the
beam splitter
hithegh-reflectivity, and rei njecti n
Fabry-Perot-typeg it, si m pl
mi y
r move
ror -i. i
e t
. into
exchange the other
the ligarm
ht using
between a
arms. Then during the next half-cycle, with arms interchanged and the
sishignft ofin thethegravi
same t y wave
direction reversed,
as each
during piece
the fi r of
st li
hal g ht
f will
-cycle. experience
At the end a phase
of the
second half-cycle, interchange the light in the two arms, and repeat as many
times as mirror losses will permit. The arrows
as a result of these exchanges, circulating around the interferometer in a in Fi g . 9.14(a) show the li g ht,
clockwise direction, but there is an equal amount of light circulating in a
countercl o ckwi s e direction. The light circulating
continual increase in phase shift throughout its circulation; the light one way experi e nces a
Fig . 9. 14. Schematic diagrams of optical configurations that could be used
for resonating light in Michelson-type (a) and Fabry-Perot-type (b) beam
receivers (Dreaver, 1983).
!fl Ell[)
E MASS
. .
. ·,
PHOTODI ODE
PHOTODIODE
(a) (b)
circulating oppositely experiences a continual decrease in phase shift; and

these oppositely
thea much
Fabry-Perot-type circulating
mirror beams,
and upon
recombining re-emerging
at the through
beam the
splitter, back
produce of
enhanced interference signal in the photodetector.
For the Fabry-Perot recei v er of Fig. 9.14(b) a single
denoted fJIR, produces the resonant interchange of the light between the two ' r esonating mirror' ,
arms:
resonancethe l e
atngths
a of the
frequency cavities are so adjusted as to produce in
fe very near the laser frequency; the reflectivities
each cavi t y a
fltc of the comer mirrors, and the path lengths between them and the
the coupled cavities at light frequencies fe + fR/2 - which requires
resonati n g mirror, are then so adjusted as to produce two resonant modes of
1 - fltc = 4nLJR/c ; (119a)
the laser frequency is then adjusted so that it drives the mode at fe -fR/2 ;
fe + R/2. In order to maximize the

and the gravity waves with frequency fR ' byf wiggling the end mirrors, then
upconvert
resulting photons into
gravity-wave-induced the mode at
signal at the photodetector, the reflecti v i t y
fltR of the resonating mirror must be adjusted to
1 - fltR = 2 1l - fltcfltE = 2 JjR
- o .
( l l9b)
With these optimizations the spectral density of shot noise near fR is (cf.
Giirsel et al., 1983; Meers, 1983; Vinet, 1986; Brillet and Meers, 1987)
s. (f) = 4S , [ + (1 :t r for I J - J. I 4 /, .
I
J (119c)
Asimprovement
is seen in inFig.noise9.13over(dotted either curves),
a si m pl this
e represents
Fabry-Perot an
or a enormous
recycl i n g
Fabry-Perot. However, this improvement is bought at the price of an
enormous narrow-banding of the receiver response: at frequencies f,..,fR/2
and f ,.., 2JR the noise is comparable to that in a simple Fabry-Perot; and at
f � JR /2 and f � 2JR it is far worse.
For
resonance,an optimally configured, resonating Michelson the
If-fR I � f0, is similar to that of a Fabry-Perot (equation (119c)).
sensitivy near
Neither recycling nor resonating is potentially useful in present
prototypes
storage times.because present
However, inmirror
the pl reflectivities
a nned limit
full-scale the
LIGOs prototypes
where to short
100-fol d
longer arms permit 100-fold greater storage times, recycling and resonating
both give promise of substantial improvements
sensitivities. The first attempts to implement recycling on a prototype were in beam-detector
430 K. S . Thorne
carri e d out by the Munich group i n 1986, with moderate success.

Resonating has not yet been attempted.
(f) Quantum limit, quantum non-demolition, and squeezed vacuum
Ausedbeamto recei v er is si m ilar to a ' H ei s enberg mi c roscope'
measure the position of a particle. The analog of the Heisenberg , in which a photon is
photon i s the laser beam and the analogs of the
end and corner masses. Just as the Heisenberg photon kicks the particle it isHeisenberg particle are the
trying to measure thereby enforcing the Heisenberg uncertainty principle,
!J.corner
x !J.p �masses
h/2, sothereby
fluctuations i n the beam' s li g ht pressure
enforcing a quantum limit on the sensitivity of thekick the end and
beam
thei r recei v
ultimate er. As I shall discuss below, these quantum fluctuations have
source i n the vacuum fluctuations of quantum
eleCaves
ctrodynami c s (Caves, 1980, 198 1).
( 1987b) has used Feynman path-integral techniques (Caves, 1986,
1987a) to derive a pseudospectral density of displacement noise that
characteri
recei v er' s z es the
swinging quantum
masses) limit
when ofitany
is harmonic
studi e d at oscillator
frequenci e s(such
high as a
above beam
the
resonant frequency 0, f using techniques that produce the minimum possible
noise:
2h
(f)
S x = m (2nf) 2 ' ( 120)
where m is the oscil ator's mass and x is its displacement from equilibrium.
For a1 Hzbeamfar recei v er, whose swinging masses have resonant frequencies
f 0 - bel o w the region of interest, this pseudospectral density for
their di s pl a cements can be transl a ted into a pseudospectral
gravity-wave field h(t) that one reads off the receiver's output: density for the
Sh
Sh (f) = m (2nf) 2L2 • ( 12 1)
Here each of the masses on which a mirror is attached is assumed to have the
same mass m. This pseudospectral density can be used, in the same manner
astheanstandard
ordinaryquantum
classical spectral density (subsection (c) above), to compute
limit on the performance of a beam detector in
variAsousforsituations.
a bar detector, so also for a beam detector, there
ofyetcifound
rcumventing this standard quantum limit. should be methods
Unfortunately
a remotely practical way of doing so, except in one specinobody has
Unruh ( 1982) has invented a method, which Caves' path-integral formalism a l case:
says should work (Caves, 1987b), for beating the standard quantum limit
when
waves. performing
The key i d a
ea narrow-band
(which was measurement
invented of
independently periodic and gravi
earlier t ational
i n the
classical domain by Gordienko, Gusev and Rudenko, 1977) is to place a
spring
thereby between
turning each
the mirror
mirror and
and the
compani companion
o n i n to mass
a two-mode on which system.i t ri d es,
By
setting the ratio of the spring frequency over the gravity-wave frequency
equal
the to
little the
mirror, square one root
can of the
force mass
the of
laser the
beam' big s companion
fluctuating over
li g ht the mass
pressure of
to
act on the motion of the big mass and not the mirror, while the
interf e
immunity rometer of thereads out
receiver the
to mirror'
Heisenberg' s motion.
s quantumThe result
noi s e is
overan i
a m proved
narrow
frequency
whether i t band.
can be Whi l e this
implemented scheme i n ai s clever
practi c aland conceptually
manner i s not satisfactory,
yet clear
Although, for broad-band measurements with beam receivers, there as yet .
is no known way of beating the standard quantum limit, Caves ( 198 1) has
invented a clever scheme for getting closer
inadequate laser power causes the shot noise to exceed it. The key to Caves' to the quantum li m i t when
idea is his discovery that the ultimate source. of the photon shot noise and the
fluctuating radiation-pressure noise i s not, as people previously thought,
fluctuations
electrodynami i n c the laser
vacuum output.
state (' v Rather,
acuum i t i s fluctuations
fluctuations' ) that in the
enter quantum
the beam
splitter
light as at
it ri
headsg ht-angles
toward to
thethe
twoincoming
arms. As laser
Caves light hasandshown,superpose by ' s on
queezithe n laser
g the
vacuum' (i.e. reducing the vacuum fluctuations in the cos[(ct-x)/...te] part of
the light while increasi n g them in the sin[(ct -x)/ ...te] part; achievable in
princi p le by sending the vacuum through a
one can reduce the beam receiver's shot noise at the expense of increasing the pumped, non-linear medium),
fluctuations in its light-pressure noise. The net result is the same as if one
were using a laser of hi g her power: in
power is too low to permit achieving the quantum limit, one improves thethe typi c al case, where the actual
sensitivity
Recentlyand Cavesmoves ( 1987toward the quantum limit.
c) has elucidated the ultimate sensitivities achievable
when one combines hi s squeezed-vacuum techni
resonating. He finds that because, in a resonating system, noise associated q ue wi t h light recycling and
with losses i n the end mirrors is as important
fluctuations entering the beam splitter, squeezing of the vacuum is not on resonance as vacuum
useful. By contrast, when combined with light recycling, squeezed-vacuum
techniques can reduce Sh(f), over a broad band of frequencies llf � f, to S0
432 K . S. Thorne
or to the quantum limit, whichever is larger:

o 10
·S
2hcXe 1 - � E 2
11
=
2L
( )
Sh(f) = max
8h
(122)
m(2nf) 2L2
Squeezed states of
usher et al.,quantum
(Slof non-linear light have been
1985; Wuoptiet al.cs., This produced
1986),achievement by experimenters
,triggering great excitement in recentl
the y
field
has triggered an effort by
the Muni c h group to implement Caves' squeezed-vacuum
effort that is ·likely to succeed within the next few years. technique - an
(g) Anticipated sensitivities of full-scale LI GO receivers

The present Munich burst sensitivity h 3 1 yr "" 4 x 10-17 atf"" 1000 Hz, when
scal
arm e d from
length the
of 3 prototype
km, would arm l
becomee ngth of 30 m to the planned Munich LIGO
h 31Y� "" 4 x 10 - 19. Although scaling up is
not straightforward (mirror diameters must be increased
example), the experimenters are quite confident of doing somewhat better a factor ten, for
than this
operating. scaled-up number by the early 1990s when the first LIGOs start
The upper thick, dashed curves in Fi g s. 9.4, 9. 6 and 9. 7 (earlier in this
chapter) show possible burst, periodic and stochasti c
a !-year search time and 90 % confidence levels), for a first generation ofsensitivities (based on
detectors in the full-scale LIGOs. These curves correspond,
above -500 Hz, to receivers that are shot-noise-limited, without recycling or at frequencies
resonating, with Argon-ion laser l i ght .te = 0.08 1 8 µm, with (laser power) x
(photodetector efficiency)= I017 = 10 W, and with light storage times of a
half-gravity-wave period or longer, so [cf. equation (115) and Fig. 9.12]
[Sh(f)] 1 = [ 1 e ( ) ] = (2.4 x 10 - 22 Hz - 1) (
2hc J:. nf 2 1 f )
017 c 1000 Hz ; (123a)
and, correspondingly (equations (111)-(113))
h '"' = 11 (f,S.Jl = 8 x 10 - 2 0
(i00::' Hz )1 for bursts, (123b)
h,1,, = 3.8(S, x 10-1 Hz)l= 2. 9 x 10- 2' (
100� Hz )
for periodic sources, (123c)
( 10-f7 Hz )--a(JSh)i
h 3 /y r = 4.5 .l
(123c)
= 1 . 1 10- 22( lOOOHz
x
f ) for stochastic sources, (123c)
a
where the search for stochastic waves is presumed to use a bandwidth ll.f= f.
Below
performance ""500 ofHzthese i t i s presumed
first-generation that
LIGO seismic
receivers. noise debilitates the
During the years following the first gravity-wave searches in the LIGOs,
the experimenters plan to run a sequence of detectors with ever improving
sensitivities,
seismic isolati pushing
o n so the
the sensitivities
detectors can ever
operate downward,
at ever and
decreasi improving
n g minimum the
frequencies. A reasonable goal by the end of the 1990s is to approach the
lower-most thi c k-dashed curves in Fi g s. 9. 4 , 9. 6 and 9. 7 . These curves
L =(armto length)=
correspond advanced detectors with the following characteristics:
4 km,
Jee= 0.0818 µm (Argon-ion laser light),
1 0 17=(laser power) (photodetector sensitivity)= 100 W,
x
BlE=Blmax=(maximum mirror reflectivity)=0.9999,

m=(mi r
photon ror mass)=
shot noise 1000 kg,
dominant at 100 Hz;S ::S 104 Hz,
quantum-limit noise dominant at 10 Hz::S ff ::S 100 Hz,
seismic noise dominant at f ::S 10 Hz. (124)
is presumed that light recycling is used for burst searches, with the knee
Itfrequency adjusted to equal the frequency of interest, so the photon shot
2nhJce (1 -LBlE) fc2]1
noise (equations (111), (117c) and (116b,c) with fk=fc) is
h 3 / y r = l1( fc Sh) 2 = 11(S o fc2/fo ) 2 = 11 [
.l .!_
1017
= 1.3 10- 2 1 (i [�z) for bursts;

x (125a)
and it is presumed that light resonating is used for periodic and stochastic
searches, with the resonant frequency adjusted to equal
interest, so the) is photon shot noise (equations (112), (113), (119c) and (116b,c) the frequency of
with fR =f
h 31 ,, = 3. 8 (4S, 10- 7 Hz)l = 3. s [2;::' 10- 1 Hz Jc-L.'Ji' E)
x x
=2. 1 x 10- 2 8 for periodic waves, ( 125b)

434 K. S. Thorne
h 3 yr =4 . 5 ( J, )- ± (4S f)1=4.5 [
°
h 2 c e fJ[-47t ( 1 - &lE)3 10 - 7 Hz1
J:. x
/ 10-7 Hz o /o 'IJ C L
=5.2 10 - 2 5 ( l kHz
x
f )1 for stochastic waves. (125c)
Here it is assumed that the stochasti c
bandwidth !!if=f0 over which the resonating receiver has goodsearch is restri c ted to the narrow
performance. The quantum-li m i t noise, which
levels at f� 100 Hz, is given by (equations (121) and (111)-(11 3)) exceeds these shot noise
2 h 1]
h 3 1yr = 11 [ 2 2 = 1 . 3 1 0 _ 2 3 x
( 1000 Hz)t for bursts, ( 126a)
1t mL f f
2 10- 71 Hz]
h 3/yr = 3 . 8 [tt 2 2
h t
mL f
= 4.4 10 - 29 (1cxx;. Hz ) for periodic waves,
x (126b)
h3/yr = 4.5 [f0(S10 -QL1 /S0) t]-± [J4 fSQL t
Hz J
i2 [� ( 10 - 1 Hz)2 r
=4· 5 (�3n )t(� mL
) 0c f J
/ 17
= 1. 5 10- 2 5 ( 1000f Hz )± for stochasti.c waves.

x ( 126c)
Here for stochastic waves s0L is the quantum-limit spectral density
(equation (121)), and the chosen bandwidth !!if= f0(SQL/S0i1· is that which
makes shot noise (equation (119c) i n tegrated over
to � the quantum noise and minimizes their sum. At frequencies 11f , with f0 <11 f <fR)
equal
f� 10 Hz, seismic noise is presumed to strongly debilitate the receiver
performances (cf. Saulson, 1984).
Figs. 9.4, 9.6 and 9.7 show, together with the above receiver sensitivities,
the characteri s ti c strengths of the gravi t y waves
that were discussed in Section 9.4 above, and the present and projected from a vari e ty of sources
sensitivities of other kinds of detectors. As these diagrams suggest, the
prospects for successful detection of gravi t y waves with
are high: at the sensitivities of the 'advanced detectors' one could see the the planned LIGOs
coalescence of neutron-star binaries at -i the Hubble distance where the
event rate is esti m ated to be several per day, the coalescence of 1OM0 black
hole binari9 es throughout the universe, supernovae in our galaxy if they put
out 10- M0 of energy near 1 000 Hz frequency and in the Virgo cluster if
they put out 10- 3 M millisecond pulsars whose frequencies are known in
0,
advance from el e ctromagneti c measurements

or larger, CFS-unstable neutron stars with X-ray luminosities as small as if they have ellipticities 10- 8·5
1 loo that of Seo X-1, and a stochastic background at f"' 30 Hz with nGw as
small as 10-1 0 . It seems likely to me that such sensitivities are more than
adequate
sensitivities forthansuccess;
these -waves
somewhere are likely
betweento bethosedetected
of the at
' fi more modest
rst-generation'
detectors and those of the 'advanced detectors' (Figs. 9.4 , 9.6 and 9.7).
(h) Ideas for other types of beam detectors
The
experi Michelson
m ental work and to Fabry-Perot
date has configurations,
focussed, are not the on
onl ywhipossic h b almost
le types al
of l
beam
been receivers.
pursued A number
vigorously. of
Since others
the have
LIGOs beenare concei
intendedv ed of,
to but
house have not
several
diwhich
fferenta number
receiversofsimultaneously and to have lifetimes of ;<:20 years, during
generations of receivers will
of these other configurations might one day operate in a LIGO. These other be built and operated, some
configurations include: (i) A frequency-tagged interferometer or Michelson
with overlapping beams (Drever and Weiss, 198 3 ), in which the light beam in
each
beams armof i s made
successi vto
e shi f
passest i n frequency
can overlap wi t
eachh each
other round-trip
but not i n pass,
terfere so
wi that
t h the
each
other; such an interferometer is basically a Michelson but with small,
Fabry-Perot-size mirrors. (ii) Active interferometers (Bagaev et al. , 1981;
Brillet
with and
transi tTourrenc,
ions near 1983),
the in
frequency whi c hofan
theacti v
lasere medi
li g u
ht) m (atoms
resi d es i n or
the molecules
arms (or
arm - there mi g ht be onl y one) of the interferometer. (iii) Spectroscopic
detectors (Nesterikhin, Rautian and Smirnov, 1978; Brillet and Tourrenc,
198 3 ; Borde et al. , 1983), in which laser spectroscopy is used to monitor the
frequency shifts produced by the gravitational waves.
achi Although
e ve i n
sensitivities principle
comparabl these e alternative
to those of types
the of beam
standard detectors
(Michelson can
and
Fabry-Perot)
detectors look types, practi
substantially c al
less issues
promising make the
(Brillet acti
and v e and
Tourrenc, spectroscopi
198 3); and
c
the frequency-tagged
venture a tentative verdict. detector has not yet been pursued i n suffici e nt depth to
9.5.4 O ther types of earth-laboratory detectors
Ingravity-wave
addition todetectorsbar detectors and beam detectors, a number
that could operate in an earth-bound laboratory of other types of
436 K . S. Thorne
have
promi been
s ing concei
to v
justi ed
fy of. Although
substantial none
experimentalof theseeffo has
rt, looked
some aresuffi c iently
question
marks (because they have not been pursued far
rather than discards. Because I have not looked at most of them in enough enough for a . clear verdi c t)
detai
rejects. l , I shall not venture to sort out the question marks from the total
(a) Electromagnetically coupled detectors
One
groups) type isof
a transducer
re-entrant used on bar detectors
microwave-cavity (e. g
resonator. by the Moscow
whose and
capacitance Perth is
modulated by the bar's vibrations (Section 10 ofBraginsky, Mitrofanov and
Panov, 1985; Blair, 1983). There is an obvious similarity of this to a Fabry
Perot beam receiver in which each arm is an optical cavity with length
modulated
conti n uous by the motions
sequence of of the
detector swi n ging masses.
configurations In
that fact,
leads onefromcan i
onem agine
to thea
other,
the mi by
c splitting
rowave the
cavi t bar
y into pieces
gradually and
being gradually
distorted moving
and them
expanded apart,unti with
l it
becomes the optical cavity of the Fabry-Perot (Caves, 1 978).
Thi
detectors s argument
are butleads twoto the recognition
examples of that
a bar
large detectors
class ofand beam
possible
'electromagnetically coupled detectors', in which gravitational waves drive
theor, when
motionsthe ofdetector
masses,getsandlargerelectromagneti c fields measure
than a reduced wavelength, gravity waves those motions;
drive vibrations of both the electromagnetic fields and the masses, which are
coupled together.
Other
figurations, earth-laboratory-scal
besi d es resonant e
bars electromagneti
and opti c al c ally
beams, coupled
that have con
been
considered theoretically and show some promise but have not been pursued
in a serious experimental way, include: (i) large microwave caviL, tiupconverts es in which
wal l motion, dri v en by gravitational waves wi t h k �
microwave quanta from one mode to another
frequency (Pegoraro, Picasso and Radicati, 1978 ; Caves, 1 979); (ii) optical mode of slightly higher
orwavesmicfrowave cavities with x,
L � intended for detection of high-frequency
� 104 Hz, in which the gravitational waves interact directly with the
resonating electromagnetic field to move quanta from one mode into
another, or interact directl y with a DC el e ctri
quanta at the gravity-wave frequency (Braginsky et al., 1 973); (iii) as a c or magnetic field to create
specific, much studi e d example of such
ring resonator in which circularly polarized gravitational waves a cavity: an opti c al or mi c rowave
propagating
electromagnetic orthogonal
field, to the
producing resonator
a linearly plane resonate
growing phase with the
shift circulati
in the n g
field.
This scheme, proposed by Braginsky and Menskii ( 197 1), was incorrectly
analyzed by them and by me (Box 37.6 of MTW) - a source of some
embarrassment. My error was in thinking I could cover the detector with a
single proper reference frame, when its diameter Lis larger than the reduced
wavelength A of the gravitational waves it sees. My incorrect conclusion, and
that of Braginsky and Menskii ( 1971) was a quadratically growing phase
oc h(ct/ A)(ct/Ae). The correct conclusion (Linet and Tourrenc,
shift, L\<l>is a linearly
1976) growing phase shift, L\<l> oc h(ct/Ae), which makes this
scheme no better in principle than a standard beam detector. WARNING:
The literature i s full of similarly incorrect analyses
described in this section, 9.5.4; (iv) an optical cavity filled with an isotropicof the various detectors
medi u m, or an optical fiber, i n whi c h gravity-wave-produced
optical effects such as birefringence (Iacopini et al., 1979; Vinet, 1985); (v) strains induce
detectors using the Mossbauer eff e ct (Kauff m an, 1970).
For
Grishchuk a general
( review of electromagnetically
1983), and for general analyses of them, see Tourrenc and
coupled detectors, see
Grossiord ( 1974), Tourrenc (1978) and Teissier du Cros ( 1985).
Analyses of such detectors sometimes produce
conclusions because they overlook the mundane issue of thermal noise in the wildly overoptimistic
mechanical parts of the detector (e. g . the walls
For an example of an analysis that does take thermal noise into account of an el e ctromagnetic cavity).
properly, see Caves ( 1979).
(b) Superfluid interferometers and superconducting circuits
Anandan ( 198 1), Chiao ( 1982) and Anandan and Chiao ( 1982) have
suggested
toarethebased, that if
superconducting it is possible
weak to
l i construct
nk (Josephson a superfluid
junction) weak
on li n
whichk, anal o
SQUIDSgous
then such a link could form the basis for a superfluid ring
interferometer that would be sens1 t
Unfortunately, such weak links do not yet exist. Schrader ( 1984) and 1ve to gravitational waves.
independently Anandan ( 1985, 1986) have suggested a gravity-wave
detector
thechange
magnetic based on
field the
in direct interaction
superconducting between
solenoids, the
withgravitational
the resulting wave and
current
monitored by a SQUID.
9.5.5 Low-frequency detectors ( 1 0-10 - 5 Hz)
As one goes to lower and lower frequencies it becomes harder and harder to
438 K. S. Thorne .
isolate
and a gravity-wave
acoustic vi b rations detector
and i n
from an earth-bourid
fluctuating laboratory
gravity gradients from seismic
due to
people, animals, trucks, etc. Ultimately, somewhere around 10 Hz, isolation
will become impossibly difficult (see, e. g .
the 'low-frequency region' 10 Hz}>f�10 - 5 Hz will require putting , Saul son, 1984 ). Thus, to operate in
detectors
detectors. in space, or using normal modes of the earth or the sun as the
been A number
conceived of space-borne
of for operati oro nearth-
at low or sun-normal-mode
frequencies; and detectors
several have have
been
constructed and have produced
waves. In this section I shall describe these briefly. interesting limits on cosmic gravitational
(a) Doppler tracking of spacecraft
Atwaveperiods
detector between
is the a few
doppler minutes
tracki and
n g a
of few hours
spacecraft. the best present gravity
used In doppler
to control tracki
the n g a highly
frequency stable
of a clock
monochromaticon earth (' m
radi aster
o oscillator'
wave, which ) is
is
transmi t ted from
byusedthebyspacecraft earth
and ' t to the spacecraft.
ransponded' back This
to the' u plink'
earth; radi
i. e o
thewaveupl i is
nk recei
wavev edi s
the spacecraft to control, in a phase-coherent way, the frequency of .
the 'downlink' radio wave that it transmits back to earth. When the down
link
theWhen radi o wave
mastergravitationali s recei v ed
oscillator; andradiation at earth,
from thatsweeps its frequency
comparison a i s compared
doppler shi ft wi
i s t h that
read outof
through the solar system - most .
idinsteresti
tance n gly
so it with
must a
be reduced
analyzed wavelength
by TT methods of order
-it the
perturbs earth-spacecraft
the earth, the
spacecraft, and the propagating radio wave.
fluctuations in the measured doppler shift with magnitude h ],!'. Each The net result is to
c5v/v ,...,
produce
feature i n the gravitational waveform hjkT ( t ) shows up three times in the
doppler
radio waveshift, in
with a manner
the gravitythat can
wave beat regarded
the events as due
of to
emission interactionfrom of the
earth,
transpondi n g by spacecraft, and reception
Wahlquist, 1975). This triplet structure can be used to help distinguish the at earth (Estabrook and
effects of the gravitational wave from noise in the doppler data.
The use of doppler data for gravi t y-wave
Braginsky and Gertsenshtein ( 1967), and was first pursued with preexisting searches was fi r st proposed by
data by Anderson
was worked(1975) (1971).
out byforDavies The response
(1974) for of doppler
special casestracki
and n g
by to a gravity
Estabrook wave
and
Wahlquist the general case. Experimental feasibility and noise
thresholds were first discussed by Wahlquist et al. (1977). Experimental

results have been obtained wi t h the Viking spacecraft (Armstrong et al.,
1979), the Voyager spacecraft (Hellings et al., 1981), Pioneer 10 (Anderson
et al. , 1984), and Pioneer 11 (Armstrong et al., 1987). A future experiment
will entail coordinated observations with the Galileo and Ulysses spacecraft
(Estabrook,
In al l past 1987).
experiments the most serious noise source was fluctuations in
the interplanetary plasma (solar wind), which cause fluctuations in the
dispersion of the radio waves and thence fluctuations in the doppler shift .
These fluctuations have li m ited the sensitivity under
[fSh(f)]1 ,...., 3 10-14 in the band 10- 4 Hz:S f:S 10 2 Hz. Fortunately,
todispersion x
optimal
-
circumstances
higher in an
radio-wave ioni z ed plasma
frequencies - becomes
a move less
that serious
is also when
desired one to moves
produceto
higher
Some bit
future rates for
spacecraft, information
including transfer
the between
Galileo the
mission, spacecraft
wil l use and
X-band earth.
radio
signals (10 GHz frequency) on both the uplink and the downlink, rather
than S-band
correspondingly (2 GHz)
thei r up and
gravity-wave X-band down
sensitivities as in
should the past;
improve and
to
[fSh(f)] L-., 3 10-15. In fact, on such spacecraft the ionized-plasma
x
dispersion
vapor in will
the be
earth' sufficiently
s atmosphere low that
may fluctuations
become the in dispersion
most serious due
noise to water
source
(Armstrong and Sramek, 1982). Plans for monitoring the water vapor as a
means of reducing this noise are being devel o ped (Resch et al., 1984) . In the
more distant future, doppler tracking may
frequency X-band/K-band (30 GHz) tracking to reduce and monitor the be improved by using dual
ionized
spacecraft plasma
(Smarr dispersion, by installing a highly stable clock
et al., 198 3 ) and/or by moving the earth-based antenna
on board the
into earth orbit. These improvements might ultimately produce [fSh(f)]1 as
low as 10-17 (Estabrook, 1987).
Figs. 9.4, 9.6 and 9. 7 show the sensitivities h3;yr corresponding to these
noise levels (equations (34) (52a) and (67) with 1�.f f).
, =
At frequencies f- 3 Hz, far above the regime 10-2-10-4 Hz of normal

doppler tracking, but below the regi m e of earth-based
hope to,...., use doppler tracking of spacecraft in earth orbit (distances detectors, one might
c ., /4f 25 000System
,...Positioning km). (GPS).
The bestUnfortunately,
such spacecraftevenarewitthose of the
h the best projected Global
improvements of the GPS we cannot anticipate [fSh(f)]1 better than
-see10-14 corresponding to a burst sensitivity
Hansen, Chiu and Chao (1986) for a detailed study. h3;yr "' 10-13; see Fi g . 9.4 and
440 K . S. Thorne
(b) Beam detectors in space

Several conceptual designs for optical beam detectors in space were
suggested by Weiss et al,. ( 1976) and are discussed in Weiss· ( 1979). More
recentl y Faller and Bender ( 198 1), Bender et al. ( 1984) and Faller et al.
( 1985) have been carrying out a detailed design and feasibility study for such
detectors; and it is possible they will fly sometime around the turn of the
century. As now concei v ed (Faller et al., 1985), such a detector might consist
ofThethree ' d rag-free' satellites, one at the comer and
satellites might be in the same solar orbit as the earth-moon system, buttwo at the ends of an L.
farkm.from the earth and moon; and their arm l e ngth might be L - 1 million
Each satelli t e might carry its own 100 mW laser, with the end lasers
phase locked to the light that arri v es from the corner laser. Wi t h 50 cm
diameter telescopes on each satelli t e to focus the laser beams, this could
result in a one-pass (B = 1) Michelson interferometer system with shot
noise-limited sensitivity [Sh(f)J t = 1 0- 2 0 Hz - t over the range 1 0 - 1 Hz �
f <, 1 0 - 4 Hz. Above 10- 1 Hz the sensitivity would be degraded
because of too-long an arm length (L > A) . Below 10 - 4 Hz stochastic non
gravi
transl taati
teso nal
into forces
the might
amplitude degrade
levels the sensitivity. This projected sensitivity
h3;yr (equations (34), (52a) and (67) with
=
df f) shown i n Fig s. 9.4, 9.6 and 9.7 .
(c) Earth's normal modes
Forward et al. ( 196 1) and Weber ( 1967) pioneered the use of the earth's
quadrupolar normal modes as 'resonant-bar' gravity-wave detectors. More
recentl y Boughn and Kuhn ( 1984) have gi v en a careful formulati o n of the
method for i n ferri n g, from earth-normal-mode
stochastic gravitational waves that might be exciting the earth. Using their observations, limits on any
method on data from the IDA sei s mi c network, Boughn and Kuhn ( 1987)
have derived a slightly stronger limit on stochastic background than the
earl y Weber ( 1967) limit: they find OGw (f) ::S 1 at f = 0.3 1 mHz (Fig . 9.7).
They expect,
lowered to QGW ;S 0 further,
J.
that a larger amount of data may permi t this limit to be
(d) Sun's normal modes
Walgate ( 1983) stimulated people to think about the sun's quadrupolar
normal modes as gravity-wave detectors, when he speculated (falsely; see,
e. g . , Kuhn and Bough, 1984, and Anderson et al., 1984) that an observed
mode was being excited by monochromatic gravitational waves from the
binary star Geminga.
Boughn and Kuhn (1984) have carried out a detailed analysis of the
gravity-wave
Isaac (1981) i
on m plications
low-order of observed
p-
and g-modes solar pulsations.
they find Q From
w(f)� the data
100 at f':!;
of
x
3orders10 -of4 Hz;magnitudeand theyimprovement,
argue that better observati o nal Gdata may produce
bringing QGw(f) Well below unity.
(e) Vibrations of blocks of the earth's crust
Braginsky et al. (1985) have recently pointed out that blocks of the earth's
crust with sizes 50 km� L � 70 km could function as resonant-bar detectors
forwithwaves of frequency f ':!; 0.03 Hz. They propose monitoring such a block
an array of seismic stations and cross-correlating the data to pull out
the sei s mi
calcorrespondi c motions
culationsng suggest associ a a ted with
possi b the
le block' s
sens1tlv1ty quadrupolar modes.
[fSh(f)JL 2 10 - 1 7
x
Their
to h3;yr 2 10 - 1 6 for bursts.
,..._, x
..
.
(f) Skyhook
Bragi n sky and Thorne
gravity-wave detector3that10would (1985) have suggested
operate in an
the earth-orbiti n g ' s kyhook'
0. 1 --0.0 1 Hz region with
sensitivity [fSh(f)Ji ,..._, x - 1 7, which is much better than present or near
fublocks
ture doppler
of the earth' tracki
s n g and
crust (seeroughly
Fi g s. comparabl
9. 4 , 9. 6 and e9.to
7 ). that
The hoped
skyhook for from
would
consist of two masses, one on each end of a
center. As it orbits the earth, the cable would be stretched radially by thelong thin cable with a spri n g at i t s
earth's tidal gravitational field. Gravitational waves would pull the masses
apart and push them together in an osci l latory
be transmitted to the spring by the cable; and a sensor would monitor the fashi o n; thei r motion woul d
spriIfnitg'severresultiflies,ngthemotion.
skyhook's role will be to provide, with a simple and
inexpensi v e device, a moderate-sensi t i v i t y coverage of the 0. 1--0.0 1 Hz
region during the epoch before far more sensitive beam detectors are buil t
and installed in space.
9 . 5 .6 Very-low-frequency detectors (frequencies below 1o- 5 Hz)
Atarefrequencies below about 10 - 5 Hz the only sources of gravitatio nal waves
probably stochastic background from the early universe (Sections
4 . 3 (d,e,f)); and the best detectors involve use of distant astronomical
9.bodies.
442 K . S. Thorne
(a) Pulsar timing

Remarkably
come from good
the timinglimits of on very-low-frequency
pulsars (Taylor, gravitational
1987). The basic idea for pulsar
waves have
timing as a gravity-wave detector (Sazhin, 1978 ; Detweiler, 1979) is this: the
rotati
pul s ar o n
ti mofi the
n g pulsar'
compares s underlying
that clock' neutron
s ti c ki n g star
rate iswia t hi
h g
thehlytistable
c ki n g clock,
of the and
very
best atomic clocks in earth-bound laboratories. When gravity waves sweep
over the pulsar, they affect the ticking rate of the pulsar's clock relative to
clasocks elsewhere
fluctuations inin the
the universe,
pulsar ti mincluding
ing data. earth;
Similarly,and those
when eff e cts
gravity show up
waves
sweep
those over
elsewhere the earth,
in the they affect
universe; the
and ticking
those rates
effe cts of all
show our
up clocks
in pul s relative
ar to
timing.
Tocan theinferextent,
that then,
gravity that
wavesthe pulsar
of a gi v timing
en strengthdata are
are free
not of fluctuations,
sweepi n g over one
either
the pulsar or the earth. (See Blandford, Narayan and Romani, 1984, for a
discussion
data and theof how
pitfalls to extract
one the
encounters gravity-wave
i n doing i
so. n formation
) If fluctuationsfrom the
are timing
seen i n
thepulsar,timingwe probably
data and willare never actuallybe able
due toto confirm
gravity waves that sweepi
the source n g over
is the
gravity
waves rather than fluctuations in the neutron star's rotatio�. However, if
fluctuations are seen and are due to gravity waves sweeping over the earth,
wepropagation,
might be polarization
able to learnandtheirwavecauseformandby study the
simultaneously waves' direction
timing of
several
pulsars on different parts of the sky.
In response to the Sazhin-Detwei l er idea, Hellings and Downs ( 1983) and
Romani and Taylor ( 1983) used timing data on four especially quiet pulsars
(which Downs and Reichley ( 1983) had tracked for over a decade with the
Goldstone antenna of 3NASA' s deep space tracki n g network) to place a
mit flGw (f)in� the1 frequency
libackground x 10- (//10 Hz)4 (90 % confidence) on any stochastic
-8
range 4 10 - 9 � f� 10 - 7 Hz; Bertotti, Carr

x
and Rees ( 1983) used data from Helfand et al. ( 1980) to obtain a similar
limit.Intrinsic noise in all the pulsars used in these analyses would have made it
difficult to improve on these limits at fixed frequency (though, by further
observations the region covered could have been extended to lower
frequencies; the lowest being of order l/(total time since the measurements
began)). Fortunately, while these analyses were in process, they were put out
ofanybusiness by the discovery (Backer et al., 1982) of a pulsar far quieter than
previously known: the millisecond pulsar PSR 1937 + 2 1 . From three
years of 1937 + 2 1 timing data taken with the Arecibo radio telescope, Davis
et al. ( 1985) and Taylor ( 1987) have now placed the limit
QGw (f) :::s; 1 x 10 -
6
( f
1 0 _ 8 Hz
)4
for f '1::, fmin = 1 0 - s Hz ( 127)
on any isotropic stochastic background : see Fig . 9. 7 .
This level of sensitivity is so good that further progress at fixed frequency
is limited by the long-term frequency stability of the world's best atomic
clocks. Thus, unless clocks improve, we can expect the coefficient in ( 127) to
improve at best as C 1 , while the lower frequency limit fmin decreases as C 1
producing !l G w ( fmin ) oc C 5 (with t = O in 1982). Ultimately, when clocks
have improved by one or two orders of magnitude, noise due to interstellar
scintillation may become a problem (Armstrong , 1984).
Recently two other quiet, fast pulsars PSR 1855 + 09 and PSR 1953 + 29
have been discovered (Segelstein et al., 1986). Together with PSR 1937 + 2 1
and others that we can hope for, they may one day form a network for
gravitational-wave searches and observations. Such a network would
alleviate problems with interstellar scintillations and atomic clock
fluctuations.
(b) The timing of orbital motions

A gravitational wave with period long compared to observation times will
produce a gradual secular change in the relative ticking rates of clocks in and
out of the wave : w/w ,..., ( f/2 n) h , where f is the gravity-wave frequency, h is
its amplitude, and w is the ticking rate of the clock in the wave as monitored
by any clock that is outside of the wave. For stochastic waves this gives
w/w "" [n Gw ( f)]1 ( 10 1 0 yr) - 1 . Because the physical torques on pulsars are so
large (w/w � ( l01 0 yr) - 1 ), pulsars cannot be useful in searches for waves
with periods longer than the observation time. However (Mashhoon, Carr
and Hu, 198 1 ; Bertotti, Carr and Rees, 1983), the orbit of the binary pulsar is
such a good clock, with such a well-understood slow down, that it can be
beat against earth-based clocks to give interesting limits on waves with
frequencies 10 - s Hz ;S / :$ 10 - 1 3 Hz. (The low-frequency limit, 10 - 1 3 Hz,
corresponds to waves with � of order the distance between the earth and the
pulsar.) The most recent observational data from Weisberg and Taylor
( 1984) give the limit
110-13Hz
10 -6Hz Q G w ( f)f - 1 df � 0 .5H i°lo ,
where H 00 is the Hubble constant in units of 100 km/s Mpc - 1 .

( 128)
1
444 K. S. Thorne
Orbital motions are also useful as detectors of waves with frequencies f of

order the orbital frequency. The orbiting bodies respond to such waves in
the same manner as does a resonant-bar detector. (After all, the orbit is, in a
sense, nothing but a two-dimensional oscillator.) The idea of using a lunar
or planetary or binary-star orbit in this way as a gravity-wave detector has
been proposed or discussed by Braginsky and Gertsenshtein (1967),
Anderson (1971), Bertotti (1973), Rudenko (1975) and many others.
Mashhoon, Carr and Hu (1981) argue that Viking doppler data on the
relative orbits of Mars and Earth place a limit nGw(f);:S0.1 at f �
3 x 10- 8 Hz (period of one year), and that laser ranging data on the moon's
orbit - which now giveQGw(f)::S 10 - have the potential in some years to
give nGw(f):SO.l at f�3 x 10 7 Hz (period of one month).
-
(c) Anisotropies in the temperature of the cosmic microwave radiation

Large-scale (quadrupolar) anisotropies in the cosmic microwave radiation
would be produced by gravitational waves in the vicinity of the earth today
(Sachs and Wolfe, 1967; Dautcourt, 1969). Present observational limits on
such anisotropies imply nGw(f)::S 10-6(f/3 x 10-1 8 Hz)2 for any f, even
f ;:S 1/(Hubble time) (Grishchuk and Zel'dovich, 1978). Small-scale
anisotropies (angular scales of order 5 degrees) would have been produced
by gravitational waves during the epoch of plasma recombination, when the
microwave radiation we now see last interacted with matter. The limit from
these is remarkably good: nGw(f) ;:S 10-1 3 in a narrow frequency window at
f �10-16 Hz (Carr, 1980; Zel'dovich and Novikov, 1983; Starobinsky,
1985); but this limit is much less firmly based than others. It presumes we
understand thoroughly the propagation of the cosmic microwave radiation
during recombination and from the epoch of recombination to the present.
Note: this limit does not constrain waves from cosmic strings or any other
source that emitted most of its waves later than the epoch of recombination;
see Traschen, Turok and Brandenberger (1986) for the effects of strings on
the microwave anisotropy.
(d) Other astronomical observations

In addition to the timing of pulsars and of orbital motions, and microwave
anisotropies, several other astronomical observations can be used as probes
of low-frequency gravitational waves: (i) the observed velocities of galaxies
and clusters of galaxies, i.e. deviations from the Hubble flow (Rees, 1971;
Burke, 1975; Dautcourt, 1977). These give QGw(f) ;:S (f/3 x 10-17)2 for
10-17 Hz;:Sf:S10-15Hz (Carr, 1980). (ii) Peculiarities in primordial
nucleosynthesis, produced by the influence of gravitational wave energy on
the expansion rate of the universe during nucleosynthesis (the 'first three
minutes') (Schvartzman , 1969). These produce a limit !lGw(f) � 10 - 4 at

f <: 1 0 - 1 0 Hz, which is not fully firmly based because it presumes we
thoroughly understand conditions in the first few minutes of the universe.
Note that this limit is irrelevant for waves emitted since the epoch of
nucleosynthesis.
For detailed reviews and references on these and other 'astronomical
detectors' of gravity waves, see Dautcourt ( 1974), Chapter 17 of Zel'dovich
and Novikov ( 1983), and Section 4 of Carr ( 1 980).
9.6 Conclusion
As I look back over this review of gravitational waves, I am struck by the
enormous changes in our theoretical understanding - or at least in
theoretical fashion - that have occurred over the past 5 , 10 and 1 5 years ; and
I am impressed even more by the progress that experimenters have made in
the quest to invent, design and build detectors of ever greater sensitivity.
That the quest ultimately will succeed seems almost assured. The only
question is when, and with how much further effort. Five years ago Jerry
Ostriker and I made a bet , to wit :
Whereas both Jeremiah P. Ostriker and K ip S. Thorne believe that
Einstein's equations are valid
And both are convinced that these equations predict the existence of
gravitational waves
And both are confident that Nature will provide what physical law
predicts
And both have faith that scientists can ultimately observe whatever
Nature does supply
Nevertheless, they differ on the likely strengths of natural sources and on
the probability of a near-future and verifiable detection.
Therefore they agree to wager one case of good red wine (JPO to supply
French wine, KST to supply California) on the detection of
extraterrestrial gravitational waves before the next Millennium (January
1 , 2000). KST wins the wager if at least two experimental groups observe
phenomena which they agree are gravitational waves. If not, JPO wins.
Signed and officially sealed
this sixth day of May 198 1
Jeremiah P. Ostriker
Kip S. Thorne
I expect to win - but I won't guarantee it.
446 K . S. Thorne
Acknowledgments
For helpful comments on the manuscript of this chapter thank Thibault
Damour, Ron Drever, John Armstrong , Peter Bender, Jin Bicak , David
I
Blair, Roger Blandford, Herman Bondi , Carlton Caves, Frank Estabrook ,
Charles Evans, Sam Finn , Craig Hogan, Jim Hough, Ed Leaver, Brian
Meers, Roger Romani, David Schoemaker, Dan Stinebring, and Kimio
Tsubono.
References
Abbott, L. and Wise, M . ( 1 984). Nuclear Physics B, 244, 54 1 .

Alpar, M . A . and Pines, D . ( 1985) . Nature, 314, 334.
Amaldi, E. and Pizella, G. ( 1979). In Relativity, Quanta, and Cosmology in the
Development of the Scientific Thought of Einstein, Volume 1 , 9.
Amaldi, E., Coccia, E., Cosmelli, C., Ogawa, Y., Pizzella, G . , Rapagnani, P . , Ricci, F.,
Bonifazi , P . , Castellano , M. G . , Vannaroni , G., Bronzoni, F., Carelli, P., Foglietti , V . ,
Cavallari, G., Habel, R., M odena, I . and Pallottino, G. V . ( 1984). Jl Nuovo Cimento,
7C, 338.
Anandan, J. ( 198 1). Physical Review Letters, 47, 463 .
Anandan, J. ( 198 5). Physics Letters, 105A, 280.
Anandan, J. ( 1 986). In Ruffini, ed. ( 1 986).
Anandan , J. and Chiao, R . Y. ( 1982). General Relativity and Gravitation , 14, 5 1 5.
Anderson, A . J. ( 197 1). Nature, 229, 547.
Anderson, J. D . , Armstrong, J. W., Estabrook, F. B., H ellings, R. W . , Law ; E. K . and
Wahlquist, H . D. ( 1984). Nature, 308, 158.
Anderson, J . L . and Hobill , D . W. ( 1986) . I n Dynamical Spacetimes and Numerical
Relativity, ed . J. M . Centrella, p. 389. Cambridge University Press.
Anderson, J. L., Kates, R. E . , Kegeles, L . S . and M adonna, R. G . ( 1982). Physical
Review D, 25, 2038.
Armstrong, J. W . ( 1984). Nature, 307, 527.
Armstrong, J. W., Estabrook, F. B. and Wahlquist, H. D. '( 1 987). Astrophysical Journal
(in press).
Armstrong, J. W . and Sramek, R. A. ( 1982) . Radio Science, 1 7, 1 579.
Armstrong, J. W., Woo, R. and Estabrook, F. B. ( 1979). Astrophysical Journal, 230, 570.
Ashtekar, A . ( 198 3). In Deruelle and Piran, eds. ( 198 3), p. 42 1 .
Ashtekar, A . ( 1984) . In General Relativity and Gravitation, ed . B . Bertotti et al., p . 37.
Reidel : Dord recht.
Backer, D. C . , Kulkarni , S. R., Heilis, C., Davis, M . M . and Gross, W . M . ( 1982).
Nature, 300, 6 15 .
Bagaev, S . N ., Chebotaev, V. P . , Dychkov, A . S . and Goldort, V. G . ( 198 1). Applied
Physics, 25, 1 6 1 .
Bardeen, J. M . and Press, W . H . ( 1973). Journal of Mathematical Physics, 14, 7 .
Bardeen, J. M . and Piran, T . ( 1 98 3). Physics Reports, 96, 205.
Begelman, M. C., Blandford , R. D. and Rees, M. J. ( 1 980). Nature, 287, 307 .
Bender, P . L . , Faller, J. E . , H all, J. L . , Hils, D . and Vincent, M . A . ( 1 984). Abstract for
Fifth International Laser Ranging Instrumentation Workshop; Herstmonceaux, 10- 14
September, 1984.
Bendat, J. S. ( 1958). Principles and Applications of Random Noise Theory, Wiley :

New York .
Bertotti, B. ( 1973). Astrophysical Letters, 14, 5 1 .
Bertotti , B., Carr, B. J. and Rees, M . J . ( 1983). Monthly Notices of the Royal
Astronomical Society, 203, 945.
Bethe, H . ( 1986) . In Highlights in Modern Astrophysics, eds . S . L. Shapiro and S . A.
Teukolsky , p. 45. Wiley : New York.
Bicak , J. ( 1968). Proceedings of the Royal Society of London A , 302, 20 1 .
Bicak , J . ( 1985). In Galaxies, Axisymmetric Systems, and Relativity, ed . M . A. H .
M acCallum , Cambridge University Press.
Bicak, J., Hoenselaers, C. and Schmidt, B. G. ( 1983). Proceedings of the Royal Society
of London A, 390, 4 1 1 .
Blair, D . ( 1982). Physics Letters A, 91, 1 97 .
Blair, D . G . ( 1983). In Deruelle and Piran ( 1983), p. 339.
Blanchet, L. ( 1987a) . Proceedings of the Royal Society of London A, 409, 383 .
Blanchet, L . ( 1 987b) . Paper in preparation .
Blanchet, L. and Damour, T. ( 1984). Physics Letters, 104A, 82.
Blanchet, L. and Damour, T. ( 1986). Philosophical Transactions of the Royal Society (in
press).
Blandford, R. D. ( 1 979). In Smarr, ed. ( 1979), p. 1 9 1 .
Blandford , R . D . ( 1 984). Private communication .
Blandford , R . D . , Narayan , R . and Romani, R . W. ( 1984). Journal of Astrophysics and
Astronomy, 5, 369.
Blandford , R. D. and Thome, K. S. ( 1979). In General Relativity: An Einstein Centenary
Survey, eds. S. W. Hawking and W. Israel , Cambridge University Press.
Blinnikov, S. I . , Novikov, I . D . , Perevodchikova, T. V. and Polnarev, A . G . ( 1 984).
Soviet Astronomy Letters, 10, 177.
Bocko , M. F. and Johnson, W. W. ( 1984). Physical Review A, 30, 2 13 5 .
Bond, J . R . and Carr, B. J. ( 1 984) . Monthly Notices of the Royal Astronomical Society,
207, 585.
Bondi , H. ( 1957). Nature, 1 79, 1072 .
Bondi, H . ( 1960) . Nature, 186, 5 3 5 .
Bondi , H . , Pirani, F. A. E. and Robinson, I . ( 1959). Proceedings of the Royal Society of
London A, 251, 5 19.
Bondi , H . , van der Burg, M. G . J . and Metzner, A. W . K . ( 1962). Proceedings of the
Royal Society of London A, 269, 2 1 .
Bonnor, W . B. ( 1 959). Philosophical Transactions of the Royal Society of London A , 251,
233.
Bontz, R . J. and H augan, M . P . ( 19 8 1 ) . Astrophysics and Space Space Science, 78, 204.
Bontz, R . J . and Price, R . H . ( 1 979). Astrophysical Journal, 228, 560.
Borde, C . J., Sharma, J . , Tourrenc, P. and Damour, T. ( 1 983). Journal de Physique-
Lettres, 44, L98 3 .
Boughn, S. P. and Kuhn , J. R . ( 1984). Astrophysical Journal, 286, 387.
Boughn, S . P. and Kuhn, J. R . ( 1987). Paper in preparation .
Boughn, S . P., Fairbank, W. M . , Giffard, R. P . , Hollenhorst, J. N . , Mapoles, E. R.,
McAshan, M. S . , Michelson , P. F., Paik, H . J. and Taber, R . C. ( 1982). Astrophysical
Journal, 261, L 19.
Braginsky, V. B . ( 1965). Uspekhi Fizicheskikh Nauk, 86, 433.
Braginsky, V . B . ( 1983). In Deruelle and Piran ( 1983), p. 387.
Braginsky, V. B . and Gertsenshtein, M . E. ( 1967). Soviet Physics - JETP Letters, 5, 287.
448 K. S. Thorne
Braginsky, V. B. and Grishchuk, L. P. ( 1985). Zhurnal Eksperimentalnoi i Teoreticheskoi

Fiziki, 89, 744. English translation : Soviet Physics - JE TP, 62, 427.
Braginsky, V. B. and Menskii, M. B. ( 197 1 ). Soviet Physics - JETP Letters, 13, 4 17.
Braginsky, V. B. and Nazarenko, V. S . ( 197 1). In Proceedings of the Conference on
Experimental Tests of Gravitat�onal Theories, November 1 1-1 3 , 1 9 70, California
Institute of Technology, ed . R. W. Davies. Jet Propulsion Laboratory Technical
Memorandum, 33-499 : Pasadena, California .
Braginsky, V. B., Grishchuk, L. P . , Doroshkevich, A . G., Zel'dovich , Ya. B., Novikov,
I. D. and Sazhin, M. V. ( 1973). Soviet Physics - JETP, 38, 865.
Braginsky, V. B., Gusev, A . V . , M itrofanov, V. P . , Rudenko, V . N . and Yakimov, V. N.
( 1985). Uspekhi Fizicheskikh Nauk, 147, 422 .
Braginsky, V . B., Mitrofanov, V . P . and Panov, V . I . ( 1 985). Systems with Small
Dissipation. University of Chicago Press : Chicago.
Braginsky, V . B. and Thorne, K. S . ( 1985). Nature, 316, 6 10.
Braginsky, V . B. and Thorne, K. S . ( 1987). Nature (in press).
Braginsky, V. B. and Vorontsov, Yu. I. (1974). Soviet Physics-Uspekhi, 1 7, 644.
Braginsky, V. B., Vorontsov, Yu . I . and Khalili, F. Ya. ( 1978). Soviet Physics - JETP
Letters, 27, 276.
Braginsky, V. B., Vorontsov, Yu. I. and Thorne, K . S. ( 1980). Science, 209, 547.
Brill, D . R. and Hartle, J . B . ( 1964). Phys. Rev. B, 135, 27 1 .
Brillet, A . ( 1985). Anna/es de Physique, 10, 2 19.
Brillet, A. and Meers, B. ( 1987). Paper in preparation.
Brillet, A. and Tourrenc, P. ( 1983). In Deruelle and Piran ( 1983).
Burke, W. L. ( 1969). Unpublished Ph.D. thesis, Caltech.
Burke, W. L. ( 1975). Astrophysical Journal, 196, 329.
Burke, W. L . ( 1979). In Isolated Gravitating Systems in General Relativity, ed . J. Ehlers,
p. 220. North Holland : Amsterdam.
Carr, B. J. ( 1 980). Astronomy and Astrophysics, 89, 6.
Carr, B. J . ( 1986). In Inner Space/Outer Space, ed . E. W . Kolb, University of Chicago
Press : Chicago.
Carter, B. and Quintana, H. ( 1977). Physical Review D, 16, 2928 .
Caves, C. M . ( 1978). Private communication .
Caves, C. M . ( 1979). Physics Letters, SOB, 323.
Caves, C. M. ( 1980). Physical Review Letters, 45, 75.
Caves, C. M. ( 198 1). Physical Review D, 23, 1693 .
Caves, C. M . ( 1982). Physical Review D, 26, 1 8 17.
Caves, C. M. ( 1983). Foundations of Quantum Mechanics, ed . S. Kamefuchi et al. , p . 195.
Physical Society of Japan : Tokyo.
Caves, C. M . ( 1986). Physical Review D, 33, 1643.
Caves, C. M. ( 1987a) . Physical Review D, 35, 18 15.
Caves, C. N. ( 1987b). I n Quantum Measurement and Chaos, ed . E. R . Pike. Plenum : Ne\\
York.
Caves, C. M . ( 1987c). Paper in preparation .
Caves, C. M . , Thorne, K . S., Drever, R. W. P . , Sandberg, V . D . and Zimmermann, M .
( 1980). Reviews of Modern Physics, 52, 34 1 .
Chand rasekhar, S . ( 1950). Radiative Transfer. Oxford University Press : London .
Chandrasekhar, S . ( 1970). Physical Review Letters, 24, 6 1 1 .
Chandrasekhar, S . ( 1975). Proceedings of the Royal Society of London A , 343, 289.
Chandrasekhar, S. ( 1983). The Mathematical Theory of Black Holes. Oxford University
Press : Oxford .
Chandrasekhar, S. and Detweiler, S. L. ( 1975). Proceedings of the Royal Society of

London, 344, 44 1 .
Chandrasekhar, S . and Detweiler, S . L . ( 1976). Proceedings of the Royal Society of
London A , 350, 165.
Chandrasekhar, S . and Esposito, F. P. ( 1970). Astrophysical Journal, 160, 153.
Chandrasekhar, S . and Friedman , J . L . ( 1972). Astrophysical Journal, 176, 745.
Chandrasekhar, S . and Friedman, J . L . ( 1973). Astrophysical Journal, 181, 48 1 .
Chandrasekhar, S . and Xanthopoulos, B . C . ( 1986). Proceedings of the Royal Society of
London, 408, 175.
Chiao, R . Y . ( 1982). Physical Review B, 25, 1655.
Christodoulou, D . and Schmidt, B. G. ( 1979) . Commun . Math. Phys., 68, 275.
Chrzanowski , P . L . ( 1975). Physical Review D, 1 1, 2042.
Clark, J. P . A., van den Heuvel, E. P . J. and Sutantyo, W. ( 1979). Astronomy and
Astrophysics, 72, 120.
Clark, J . P. A. and Eardley, D. M. ( 1977). Astrophysical Journal, 215, 3 1 5.
Cohen , J . M . and Kegeles, L . S. ( 1975). Physics Letters, 54A, 5.
Cole, J . D . ( 1968). Perturbation Methods in Applied Mathematics. Blaisdell : Waltham,
Mass.
Crowley, R . J. and Thorne, K . S. ( 1977). Astrophysical Journal, 215, 624.
Cunningham, C. T., Price, R . H . and Moncrief, V. ( 1979). Astrophysical Journal, 230,
8 70.
Cutler, C. and Lindblom, L . ( 1987). Astrophysical Journal (in press).
Cyranski, J . F. and Lubkin, E. ( 1974) . Annals of Physics, 87, 205 .
Damou r, T . ( 1983). In Deruelle and Piran ( 1983). p. 59.
Damour, T. ( 1986). In Proceedings of the Fourth Marcel Grossman Meeting on the
R ecent Developments of General R elativity, ed . R. Ruffini. North Holland : Amsterdam
Damour, T. ( 1987). In Gravitation in Astrophysics, eds. B. Carter and J. B. Hartle.
'
Plenu m : New York .

Damour, T . and Deruelle, N . ( 1986). Anna/es Institut Henri Poincare (Physique
Theorique), 44, 263.
Dautcourt , G. ( 1969). Monthly Notices of the Royal Astronomical Society, 144, 255.
Dautcourt, G. ( 1974). In Confrontation of Cosmological Theories with Observational
Data, Proceedings of IAU Symposium 63, ed. M. S. Longair, p. 299. Reidel :
Dordrecht.
Dautcourt , G. ( 1 977). Astronomische Nachrichten, 298, 8 1 .
Davies, R . W . ( 1974). I n Col/oque Internationaux CNRS No. 220, 'Ondes et Radiutions
Gravitationelles'. Institut Henri Poincare : Paris, p. 33.
Davis, M. M . , Taylor, J . H., Weisberg, J. M. and Backer, D . C. ( 1 985). Nature, 315,
547 .
De Logi, W. K . and Kovacs, S . J . ( 1977). Physical Review D, 16, 233 1 .
Deruelle, N . and Piran, T. ( 1983). Gravitational Radiation . North Holland : Amsterdam.
de Sabbata, V. and Weber, J. eds. ( 1977). Topics in Theoretical and Experimental
Gravitation Physics. Plenum : London.
Detweiler, S. L . ( 1975). Astrophysical Journal, 197, 203.
Detweiler, S. L . ( 1977). Proceedings of the Royal Society of London A, 352, 38 1 .
Detweiler, S . L . ( 1978). Astrophysical Journal, 225, 687.
Detweiler, S . L. ( 1979). Astrophysical Journal, 234, 1 100.
Detweiler, S. L . ( 1980). Astrophysical Journal, 239, 292.
Detweiler, S . L . and Ipser, J . R . ( 1973). Astrophysical Journal, 185, 685.
Detweiler, S. L. and Lindblom, L . ( 1985). Astrophysical Journal, 292, 12.
450 K . S. Thorne
Detweiler, S . L. and Szedenits, E. ( 1979). Astrophysical Journal, 231, 2 1 1 .

DeWitt, B . S . ( 1967a). Phys. Rev., 160, 1 1 13 .
DeWitt, B . S. ( 1967b). Phys. Rev., 162, 1239 .
Douglass, D. H. and Braginsky, V. B. ( 1979). In General Relativity, an Einstein
Centenary Survey, eds. S . W . Hawking and W. Israel, Cambridge University Press.
Downs, G. S. and Reichley, P. E. ( 1983). Astrophysical Journal Supplement Series, 53,
169 .
Drever, R . W . P . ( 1977). Quarterly Journal of the Royal Astronomical Society, 18, 9.
Drever, R . W. P. ( 1983). In Deruelle and Piran ( 1983). p. 32 1 .
Drever, R . W . P . , Hough , J., Edelstein , W., Pugh, J . R . and Martin, W . ( 1977). In
Gravitazione Sperimentale, ed. B. Bertotti, p. 365. Accademia Nazionale dei Lincei :
Rome.
Drever, R. W. P . , Ford, G. M ., Hough, J., Kerr, I., Munley, A. J., Pugh, J. R . ,
Robertson, N. A. and Ward, H . ( 1980). Proceedings of the Ninth International
Conference on General Relativity and Gravitation, ed. E. Schmutzer, p. 265. VEB
Deutscher Verlag der Wissenschaften : Berlin .
Drever, R . W. P. and Weiss, R . ( 1983). Unpublished work reported briefly in Appendix
B .2(c) of D rever et al. ( 1985).
Drever, R . W. P., Weiss, R., Linsay, P . S., Saulson , P. R., Spero, R . and Schutz, B. F.
( 1985). A Detailed Engineering Design Study and Development and Testing of
Components for a Laser Interferometer Gravitational Wave Observatory. Caltech :
Pasadena, California, and MIT : Cambridge, Massachusetts (unpublished).
Dymnikova, I . G., Popov, A. K. and Zentsova, A. S. ( 1 982). Astrophysics and Space
Science, 85, 23 1 .
Dyson , F. J. ( 1969). Astrophysical Journal, 156, 529 .
Eardley, D . M . ( 1983). In Deruelle and Piran ( 1983). p . 257.
Eardley, D . M ., Lee, D . L . and Lightman, A . P . ( 1973). Physical Review D, 8, 3308 .
Eardley, D . M ., Lee, D. L., Lightman, A . P., Wagoner, R . V . and Will, C. M . ( 1973) .
Physical Review Letters, 30, 884.
Eddington, A . S. ( 1924). The Mathematical Theory of Relativity, 2nd edn, Cambridge
University Press; see especially supplementary note 8 .
Edelstein, L. A. and Vishveshwara, C. V . ( 1970). Physical Review D, 1 , 35 14.
Ehlers, J . and Kundt, W. ( 1962). In Gravitation: An Introduction to Current Research, ed.
L. Witten. Wiley : New York.
Ehlers, J . , Rosenblum, A . , Goldberg, J. N. and Havas, P. ( 1976). Astrophysical Journal
Letters, 208, L 77.
Einstein, A. ( 19 16). Preuss. Akad. Wiss. Berlin, Sitzungsberichte der physikalisch
mathematischen Klasse, p. 688 .
Einstein, A. ( 19 18). Preuss. Akad. Wiss. Berlin, Sitzungsberichte der Physikalisch-
mathematischen Klasse, p. 1 54.
Einstein, A . and Rosen, N. ( 1936). Journal of the Franklin Institute, 223, 43.
Epstein, R . and Wagoner, R . V. ( 1975). Astrophysical Journal, 197, 7 17 .
Esposito, F. P . ( 197 1a). Astrophysical Journal, 165, 165.
Esposito, F. P. ( 197 1b). Astrophysical Journal, 168, 495 .
Estabrook, F. B . ( 1985). General Relativity and Gravitation, 1 7, 7 19.
Estabrook , F . B . ( 1987). Acta Astronautica (in press).
Estabrook, F. B. and Wahlquist, H. D. ( 1975). General Relativity and Gravitation, 6, 439.
Evans, C . R . ( 1984). PhD thesis, University of Texas at Austin, unpublished.
Evans, C. R. ( 1986). In Dynamical Spacetimes and Numerical Relativity, ed. J. M .
Centrella, p . 3 . Cambridge University Press.
Evans, C. R. and Abrahams, A. M . ( 1987). Physical Review D (in preparation).

Evans, C. R., lben , I. and Smarr, L. ( 1987). Astrophysical Journal, submitted .
Faller, J . E. and Bender, P. L . ( 198 1 ) . Abstract for the International Conference on
Precision Measurements and Fundamental Constants, 8- 12 June 198 1 .
Faller, J . E., Bender, P . L., Hall, J . L., Hils, D . and Vincent, M . A. ( 1985). In
Proceedings of th e Colloquium 'Kilometric Optical Arrays in Space', Cargese (Corsica)
23-5 October 1984. ESA SP-226.
Feynman , R. P. ( 1963). Acta Physica Polonica , 24, 697 : also published in Proceedings of
the 1 962 Warsaw Conference on the Theory of Gravitation. PWN-Editions Scientifiques
de Pologne : Warsaw ( 1964) .
Finn, L . S . ( 1986). Monthly Notices of the Royal Astronomical Society, 222, 393 .
Fock, V . A. ( 1959). Theory of Space, Time, and Gravitation, Section 8 7 . Pergamon :
London .
Forward , R. L . ( 1978). Physical Review D, 17, 379.
Forward , R. L. and Moss, G. E. ( 1972). Bulletin of the American Physical Society, 17,
l 183(A).
Forward , R . L., Zipay, D., Weber, J., Smith, S. and Benioff, H. ( 196 1). Nature, 189, 473.
Friedman, J . L . , Ipser, J . R. and Parker, L . ( 1986) . Astrophysical Journal, 304, 1 1 5.
Friedman , J . L. and Schutz, B . F. ( 1975a). Astrophysical Journal (Letters), 199, L 1 57.
Friedman , J . L. and Schutz, B . F. ( 1975b). Astrophysical Journal, 200, 204.
Friedman , J. L. and Schutz, B . F. ( 1 978). Astrophysical Journal, 222, 28 1 .
Futamase, T . and Schutz, B . F . ( 1983) . Physical Review D , 28, 2363 .
Gal'tsov , D . V . , Matiukhin , A . A . and Petukhov, V . I . ( 1980) . Physics Letters, 77A, 387.
Gal'tsov, D. V., Tsvetkov, V . P . and Tsirulev, A. N. ( 1 984). Soviet Physics - JETP, 59,
472.
Geroch , R . , Held, A. and Penrose, R. ( 1973). Journal of Mathematical Physics, 14, 874.
Gertsenshtein , M . E. ( 1962). Soviet Physics - JETP, 14, 84.
Gertsenshtein , M . E. and Pustovoit, V. I. ( 1962). Soviet Physics - JETP, 16, 433.
Gilden , D. L. and Shapiro, S. L. ( 1 984). Astrophysical Journal, 287, 728.
Giffard, R . P. ( 1976). Physical Review D, 14, 2478.
Gomez, R., Isaacson, R. A . , Welling, J . S . a n d Winicour, J . ( 1986). In Dynamical
Spacetimes and Numerical Relativity, ed . J . M . Centrella, p. 236 . Cambridge University
Press .
Gordienko, N. V . , Gusev, A . V . and Rudenko, V. N . ( 1977). Vestnik Moskovskovo
Universiteta, 18, 48 .
Grishchuk, L . P . ( 1974). Soviet Physics - JETP, 40, 409.
Grishchuk , L. P. ( 1975a). Lettere al Nuovo Cimento, 12, 60.
Grishchuk, L. P. ( 1975b). Soviet Physics - JETP, 40, 409.
Grishchuk , L. P . ( 1977). Annals of the New York Academy of Sciences, 302, 439.
Grishchuk, L . P. ( 1983). In Proceedings of the Ninth International Conference on General
Relativity and Gravitation, ed. E. Schmutzer. Cambridge University Press.
Grishchuk, L. P. and Polnarev, A. G . ( 1980). fn General Relativity and Gravitation,
vol . 2, ed. A. Held, p. 393 . Plenum : New York .
Grishchuk, L. P . and Zel'dovich, Y a . B . ( 1978). Soviet Astronomy - AJ , 22, 125.
Giirsel, Y., Linsay, P., Saulson, P., Spero, R., Weiss, R. and Whitcomb, S. ( 1983).
Unpublished Caltech/MIT manuscript.
Gusev, A. V. and Rudenko, V. N. ( 1 976) . Radiotekhnika i Electronika, 3, 1865.
Haensel , P., Zdunik, J. L . and Schaeffer, R. ( 1986) . Astronomy and Astrophysics, 160,
25 1 .
Halliwell, J . J . and Hawking, S . W . ( 1985). Physical Review D , 31, 1 777.
452 K. S. Thorne
Halpern, L. E. and Desbrandes, R. ( 1969). Annales lnstitut Henri Poincare, 9, 309.

Hamilton , W. 0 . , Xu , B.-X., Solomonson, N., Mann, A . G . and Sibley, A. ( 1986).
Performance of the LSU Tuned Bar Gravitational Wave Detector. Unpublished
technical memorandum , Louisiana State University.
Hansen , P. M . , Chiu, Y. T. and Chao, C.-C. (1986). Aerospace Report No. ATR-
86(842 1)-2. The Aerospace Corporation : El Segundo, CA.
Hawking, S . W. ( 1985). Physics Letters, 1508, 339.
Haugan , M. P., Shapiro, S. L . and Wasserman , I. (1982). Astrophysical Journal, 257,
283 .
Heffner, H . (1962). Proceedings of the I RE, 50, 1604.
Helfand, D. H . , Taylor, J. H . , Backus, R . R. and Cordes, J. M . ( 1980). Astrophysical
Journal, 237, 206.
Hellings, R. W. and Downs, G. S. ( 1983). Astrophysical Journal (Letters), 265, L 39 .
Hellings, R. W., Callahan, P . S . , Anderson, J. D . and Moffett, A. T . ( 198 1). Physical
Review D, 23, 844.
Hils, D. L . , Bender, P., Faller, J. E. and Webbink, R. F. ( 1987). Paper in preparation.
Hirakawa, H. and Narihara, K . ( 1975) . Physical Review Letters, 35, 330.
Hirakawa, H., Narihara, K . and Fujimoto , M .-K. ( 1976). Journal of the Physical Society
of Japan, 41, 1093.
Hirakawa, H ., Owa, S . and Iso, K . ( 1985). Journal of the Physical Society of Japan, 54,
1270.
Hogan, C. J. (1986). Monthly Notices of the Royal Astronomical Society, 218, 629 .
Hogan, C. J . and Rees, M . J . ( 1984). Nature, 3 1 1 , 109.
Hollenhorst , J. N. (1979). Physical Review D, 19, 1669.
Hough, J . , M eers, B. J . , Newton , G. P., Robertson, N. A., Ward, H . , Schutz, B. F. and
Drever, R. W. P. ( 1986). A B ritish Long Baseline Gravitational Wave Observatory,
unpublished report submitted by Glasgow University to the S ERC.
Hough, J., Pugh, J. R., Bland , R. and Drever, R. W. P. ( 1975). Nature, 254, 498.
Hough, J., Drever, R . W. P., Munley, A . J., Lee, S.-A . , Spero , R . , Whitcomb, S. E.,
Pugh, J., Newton , G., Meers, B., Brooks, E. and Gursel , Y. ( 1 983). In Quantum
Optics, Experimental Gravity, and Measurement Theory, eds. P. Meystre and M . 0 .
Scully, p. 5 1 5. Plenum : New York.
Hu, B. L. ( 1978). Physical Review D, 18, 969.
Hu Enke, Guan Tongren, Yu Bo, Tang Mengxi, Chen Shusen, Zheng Qingzhang,
M ichelson , P. F., Moskowitz, B. E., McAshan, M . S . , Fairbank , W. M . and Bassan ,
M . ( 1986). Chinese Physics Letters No. 1 1-12, 1 .
Iacopini, E., Picasso, E., Pegoraro, F. and Radicati, L. A . ( 1 979). Physics Letters, 73A,
140.
Iben , I . and Tutukov, A. V. ( 1984). Astrophysical Journal Supplements, 54, 335.
Ipser, J . R. ( 1971). Astrophysical Journal, 166, 175.
Ipser, J. R. (1986) . Private communication.
Ipser, J . R . and M anagan , R . A . ( 1984). Astrophysical Journal, 282, 287.
Ipser, J . R. and Thorne, K . S . ( 1973). Astrophysical Journal, 181, 18 1 .
Isaac, G. R . (198 1). Solar Physics, 74, 43.
Isaacson , R. A . (1968a). Physical Review, 166, 1263 .
Isaacson, R . A. ( 1968b) . Physical Review, 166, 1272 .
Isaacson , R. A., Welling, J . S . and Winicour, J . ( 1983). Journal of Mathematical Physics,
24, 1824.
Isaacson , R. A., Welling, J . S. and Winicour, J . ( 1984). Physical Review Letters, 53, 1 870.
Kafka, P. ( 1977). In de Sabbata and Weber ( 1977), p. 16 1 .
Kahn, K . and Penrose, R . ( 197 1). Nature, 229, 185.
Kaufmann, W. J. ( 1970). Nature, 227, 157.

Kellerman, K. I. and Sheets, B . ( 1983). Serendipitous Discoveries in Radio Astronomy.
National Radio Astronomy Observatory : Green Bank , West Virginia.
Kojima, Y. and Nakamura, T. ( 1984a). Progress of Theoretical Physics, 71, 79.
"
Kojima, Y . and Nakamura, T. ( 1984b). Progress of Theoretical Physics, n, 494.
Kovacs, S. J. and Thorne, K. S. ( 1977) . Astrophysical Journal, 217, 252.
Kovacs, S . J . and Thorne, K. S . ( 1978). Astrophysical Journal, 224, 62.
Kuhn , J . R. and Boughn, S . P. ( 1984). Nature, 308, 164 .
Landau , L . D. and Lifshitz, E. M . ( 194 1). Teoriya Polya. Nauka : Moscow. (English
translation of a later edition, The Classical Theory of Fields. Addison-Wesley :
Cambridge, M ass. ( 195 1).)
Leaver, E. W. ( 1985). Solutions to a Generalized Spheroidal Wave Equation in
Molecular Physics and General Relativity, and an Analysis of the Quasi-Normal
Modes of Kerr Black Holes, unpublished PhD thesis, University of Utah : Salt Lake
City.
Leaver, E. W. ( 1986a). Proceedings of the Royal Society of wndon A, 402, 285.
Leaver, E. W. ( 1986b). Physical Review D, 34, 384.
Lifshitz, E. M . ( 1946). Zhurnal Eksperimentalnoi i Teoreticheskoi Fiziki, 16, 587.
Lindblom, L . ( 1986). Astrophysical Journal, 303, 146.
Lindblom, L. ( 1987). Astrophysical Journal (in press).
Lindblom, L . and Detweiler, S . L . ( 1983). Astrophysical Journal, 53, 73.
Linet , B . ( 1984). General Relativity and Gravitation, 16, 89.
Linet, B . and Tourrenc, P. ( 1976). Canadian Journal of Physics, 54, 1 129.
Linsay, P . , Saulson, P . , Weiss, R. and Whitcomb, S. ( 1983). A Study of a Long Baseline
Gravitational Wave Antenna System ; report prepared for The National Science
Foundation . MIT : Cambridge, M assachusetts (unpublished).
Lipunov, V . M. and Postnov , K . A . ( 1986). Soviet Astronomy Letters. In press.
Lipunov , V. M ., Postnov, K . A . and Prokhorov, X. ( 1987) . Paper in preparation .
Lyamov, V . E. and Rudenko , V . N . ( 1975). Soviet Physics - JETP, 40, 787 .
M accallum, M . A . H ., ed . ( 1987). Proceedings of the Tenth International Conference on
General Relativity and Gravitation . Cambridge University Press.
MacCallum , M . A. H . and Taub, A. H. ( 1973). Communications in Mathematical Physics,
30, 153.
M acedo, P . G. and Nelson, A . H . ( 1983). Physical Review D, 28, 2382.
Maischberger, K . , Rudiger, A., Schilling, R., Schnupp, L., Shoemaker, D. and
Winkler, W. ( 1985). Vorschlag zum Bau eines grossen Laser-Interferometers zur
M essung von Gravitationswellen . M ax-Planck Institut fur Quantenoptik : Munich
(unpublished).
Mashhoon, B., Carr, B . J . and Hu, B. L . ( 198 1). Astrophysical Journal, 246, 569 .
Matzner, R . A . , DeWitt-Morette, C . , Nelson, B. and Zhang , T.- R . (1985). Physical
Review D, 31, 1869.
Matzner, R. A. and Tipler, F. J. ( 1984) . Physical Review D, 29, 1575.
Meers, B . ( 1983). Unpublished PhD thesis, University of Glasgow.
Michelson , P. F. ( 1983). In Deruelle and Piran ( 1983), p. 465 .
M ichelson, P . F. ( 1986). Physical Review D, 34, 2966.
Michelson , P. F. and Taber, R. C. ( 198 1). Journal of Applied Physics, 52, 43 13.
Michelson, P . F. and Taber, R . C. ( 1984). Physical Review D , 29, 2 149.
Mijic, M ., Morris, M . and Suen, W .- M . (1986). Physical Review D , 34, 2934.
M ironovskii, V . N . ( 1966). Soviet Astronomy - AJ, 9, 752.
Misner, C. W., Thorne, K. S. and Wheeler, J . A. ( 1973). Gravitation. W. H. Freeman &
Co. : San Francisco. Cited in text as MTW.
454 K. S. Thorne
Moncrief, V. ( 1974). Annals of Physics, 88, 323.

Moss, G. E., Miller, L. R. and Forward, R. L . ( 197 1). Applied Optics, 10, 2495.
Miiller, E. ( 1982). Astronomy and Astrophysics, 1 14, 53.
Muller, E. ( 1984). In Problems of Collapse and Numerical Relativity, eds.' D. Bancel and
M . Signore, p. 27 1 . Reidel : Dordrecht.
Nakamura, T. ( 1983). In Deruelle and Piran ( 1983).
Nakamura, T. ( 1986). In Gravitational Collapse and Relativity, eds. H . Sato and
T. Nakamura, p. 295 . World Scientific : Singapore.
Nakamura, T. and Sasaki, M . ( 198 1). Physics Letters, 106B, 69.
Nesterikhin, Y. E., Rautian, S. G. and Smirnov, G. I . ( 1978). Soviet Physics - JETP, 48,
1.
Newman , E. T. and Penrose, R. ( 1965) . Physical Review Letters, 15, 23 1 .
Newman , E. T. and Tod , K . P . ( 1980). In General Relativity and Gravitation, Volume 2 ,
ed. A. Held, p . 1 . Plenum : New York.
Nutku, Y . and Halil, M . ( 1977) . Physical Review Letters, 39, 1379.
Oelfke, W. C. ( 1 983). In Quantum Optics, Experimental Gravitation and Measurement
Theory, eds. P. Meystre and M . 0. Scully, p. 387. Plenum : New York .
Ostriker, J . P., Thompson, C. and Witten, E. ( 1986). Physics Letters, B180, 23 1 .
Owa, S . , Fujimoto, M .-K., Hirakawa, H . , Morimoto, K . , Suzuki, T . and Tsubono, K.
( 1986). In Proceedings of Fourth Marcel Grossman Meeting on Recent Developments of
General Relativity, ed. R. Ruffini . North Holland : Amsterdam .
Paczynski, B . and Sienkiewicz, R . ( 198 1). Astrophysical Journal (Letters), 248, L27 .
Paik, H . J . and Wagoner, R. V . ( 1976). Physical R eview D, 13, 2694.
Pandharipande, V. R . , Pines, D. and Smith, R. A. ( 1976). Astrophysical Journal, 208,
550.
Pallotino, G. V. and Pizella, G. ( 1981). Nuovo Cimento, 4C, 237 .
Papadopoulos, D. and Esposito, F. P. ( 1985) . Astrophysical Journal, 282, 3 30.
Papapetrou, A. ( 1962) . Comptes Rendus Acad. Sci. Paris, 255, 1 578.
Papapetrou, A . ( 1971). Anna/es Institut Henri Poincare, 14, 79.
Papapetrou, A . ( 1977). In deSabbata and Weber ( 1 977), p. 83.
Pegoraro, F., Picasso, E. and Radicati, J. ( 1978). Journal of Physics A, 1 1, 1949.
Penrose, R. (1963a) . Physical Review Letters, 10, 66.
Penrose, R. ( 1963b). Relativity, Groups and Topology, eds. C. DeWitt and B. DeWitt,
p. 565. Gordon and Breach : New York.
Peres, A. ( 1960). Nuovo Cimento, 15, 3 5 1 .
Peters, P. C . ( 1964) . Physical Review, 136, B 1224.
Peters, P. C. and Mathews, J. ( 1963). Physical Review, 131, 435.
Piran, T. ( 1983). In Deruelle and Piran ( 1983), p . 203.
Piran, T. and Stark, R . F. ( 1986). In Dynamical Spacetimes and Numerical Relativity, ed .
J . M . Centrella, p. 40. Cambridge University Press.
Pirani, F. A. E. ( 1956). Acta Physica Polonica, 15, 389.
Pirani , F. A. E. (1957) . Physical Review, 105, 1089.
Pirani, F. A. E. ( 1964). In Lectures on General Relativity, eds. A. Trautman, F. A. E.
Pirani and H. Bondi. Prentice-Hall : Englewood Cliffs, NJ.
Power, E. A . and Wheeler, J . A . ( 1957). Reviews of Modern Physics, 29, 480.
Press, W. H. ( 1971). Astrophysical Journal (Letters), 1 70, L 105.
Press, W . H . ( 1977). Physical Review D , 15, 965 .
Press, W. H . ( 1979). General Relativity and Gravitation, 1 1, 105 .
Price, R. H . ( 1972a). Physical Review D, 5, 24 19.
Price, R. H. ( 1972b). Physical Review D , 5, 2439.
Price, R. H. and Thorne, K. S. ( 1969) . Astrophysical Journal, 155, 163.
Priedhorsky, W ., Stella, L. and White, N. E., International Astronomical Union Circular,

No. 4247, 28 August, 1986.
Quinlan , G. and Shapiro , S . L. ( 1987). Astrophysical Journal, in press.
Ramaty, R., Bonazzola, S . , Cline, T. L . , Kazanas, D. and Meszaros, P. (1980). Nature,
287, 122.
Rees, M. J . ( 197 1). Monthly Notices of the Royal Astronomical Society, 154, 187.
Rees, M. J . ( 1983). In Deruelle and Piran ( 1983), p. 297.
Regge, T. and Wheeler, J. A. ( 1957) . Physical Review, 108, 1063.
Resch, G . M . , Hogg, D. E. and Napier, P . G . ( 1 984). Radio Science, 19, 4 1 1 .
Reynolds, S . P. and Stinebring, D. R. ( 1 984). Millisecond Pulsars, National Radio
Astronomy Observatory : Green Bank, West Virginia.
Romani, R. W. and Taylor, J. H . ( 1983). Astrophysical Journal (Letters), 265, L 3 5 .
Rosen, N . ( 1937). Physikalische Zeitschift Sowjetunion, 12, 366.
Rubakov, V . , Sazhin, M. C. and Veryaskin, A. ( 1982). Physics Letters, 1 15B, 189.
Rudenko, V . N . ( 1975) . Soviet Astronomy - AJ, 19, 270.
Rudenko, V. N. and Sazhin, M . V. ( 1980). Kvantovaya Elektronika , 1, 2344.
Ruffini, R. ed. ( 1986). Proceedings of the Fourth Marcel Grossman Meeting on Recent
Developments of General Relativity. North Holland : Amsterdam.
Ruffini, R. and Wheeler, J. A. ( 1 97 1). In Proceedings of the Conference on Space Physics.
European Space Research Organization : Paris, p. 45.
Sachs, R. K . ( 1962) . Proceedings of the Royal Society of London A, 270, 103.
Sachs, R. K. ( 1963). In Relativity, Groups and Topology, eds. C. DeWitt and B. DeWitt,
p. 52 1 . Gordon and Breach : New York .
Sachs , R. K . and Wolfe, A. M . ( 1967). Astrophysical Journal, 147, 73.
Saenz, R. A. and Shapiro , S. L. ( 1 978). Astrophysical Journal, 221, 286.
Saenz, R . A. and Shapiro, S . L. ( 1979). Astrophysical Journal, 229, 1 107.
Saenz, R. A. and Shapiro , S. L. ( 198 1). Astrophysical Journal, 244, 1033.
Sasaki, M. ( 1984). In Problems of Collapse and Numerical Relativity, eds. D. Bance) and
M . Signore, p . 203 . Reidel : Dordrecht.
Sasaki, M . and Nakamura, T . ( 1982). Progress of Theoretical Physics , 67, 1788.
Sato, H. ( 1987). Paper in press. .
Saulson, P . (1 984) . Physical Review D, 30, 732.
Saulson, P. ( 1984) . Physical Review D, 30, 732 .
Sazhin, M . V . ( 1978). Soviet Astronomy - AJ, 22, 36.
Schmidt, B. G. ( 1979). In Isolated Gravitating Systems, ed. J. Ehlers, p. 1 1 . North
Holland : Amsterdam.
Schmidt , B. G. ( 1986) . Gravitational Radiation Near Spatial and Null Infinity, preprint.
Schilling , R. (1986). Private communication.
Schoemaker, D . , Winkler, W . , Maischberger, K . , Rudiger, A . , Schilling, R. and Schnupp,
L. ( 1986). In Ruffini , ed . ( 1986).
Schrader, R. ( 1984). Physics Letters B, 143B, 42 1 .
Schramm, D . N . and Arnett, W. D . ( 1975). Astrophysical Journal, 198, 629.
Schumaker, B. L. ( 1 986). Physics Reports, 135, 3 18.
Schumaker, B. L . and Thorne, K. S. ( 1983). Monthly Notices of the Royal Astronomical
Society, 203, 457.
Schutz, B. F . ( 1972). Astrophysical Journal Supplement Series, 24, 343 .
Schutz, B . F. ( 1986a). In Relativity, Supersymmetry, and Cosmology, ed . 0. Bressan,
M. Castagnino and V. Hamity, p. 3. World Scientific : Singapore .
Schutz, B . F. ( 1986b). Nature, 323, 3 10.
Schutz, B . F. ( 1986c) . In Dynamical Spacetimes and Numerical Relativity, ed . J . Centrella,
456 K. S. Thorne
Schutz, B. F. (1987). In Gravitation in Astrophysics, eds. B. Carter and J. Hartle.

Plenum: New York.
Schutz, B. F. and Tinto, M. (1987). Monthly Notices of the Royal Astron?mical Society,
224, 131.
Schvartzman, V. F. (1969). Soviet Physics - JETP Letters, 9, 184.
Segelstein, D. J., Rawley, L.A., Stinebring-, D. R., Fruchter,A. S. and Taylor, J. H.
(1986). Nature, 322, 714.
Shapiro, S. L. and Teukolsky, S. A. (1985). Astrophysical Journal (Letters), 292, L 41.
Slusher, R. E., Hollberg, L. W., Yurke, B., Mertz, J. C. and Valley, J. F. (1985). Physical
Review Letters, 55, 2409.
Smarr, L. (1977a). Annals of the New York Academy of Sciences, 302, 569.
Smarr, L. (1977b). Physical Review D, 15, 2069.
Smarr, L., ed. (1979). Sources of Gravitational Radiation. Cambridge University Press.
Smarr, L. and York, J. W. (1978). Physical Review D, 17, 2529.
Smarr, L., Vessot, R. F. C., Lundquist, C. A., Decher, R. and Piran, T. (1983).
General Relativity and Gravitation, 15, 129.
Spero, R. (1986a). In Ruffini, ed. (1986).
Spero, R. (1986b). Private communication.
Stark, R. F. and Piran, T. (1985). Physical Review Letters, 55, 891 and 56, 97.
Stark, R. F. and Piran, T. (1986). In Proceedings of the Fourth Marcel Grossman
Meeting on Recent Developments of General Relativity, ed. R. Ruffini (in press).
Starobinsky,A.A. (1979). Soviet Physics - JETP Letters, 30, 682.
Starobinsky,A. A. (1985). Soviet Astronomy Letters, 11, 133.
Stewart, J.M. and Friedrich, H. (1983). Proceedings of the Royal Society of London A,
384, 427.
Suzuki, T., Tsubono, K. and Hirakawa, H. (1978). Physics Letters A, 67, 10.
Sullivan, W. T. (1982). Classics in Radio Astronomy. Reidel: Dordrecht.
Sullivan, W. T. (1984). The Early Years of Radio Astronomy. Cambridge University
Press.
Szekeres, P. (1971). Annals of Physics, 64, 599.
Tammann, G.A. (1981). In Supernovae: A Survey of Current Research, ed. M. J. Rees
and R. J. Stoneham, p. 371. Reidel: Dordrecht.
Taylor, J. H. (1987). In General Relativity and Gravitation, ed. M.A. H.MacCallum.
Cambridge University Press (in press).
Teissier du Cros, F. (1985). Annales de Physique, 10, 263.
Taylor, J. H. (1987). In Proceedings of Eleventh International Conference on General
Relativity and Gravitation. Cambridge University Press (in press).
Teukolsky, S. A. (1972). Physical Review Letters, 29, 1114.
Teukolsky, S. A. (1973). Astrophysical Journal, 185, 635.
Teukolsky, S. A. and Press, W. H. (1974). Astrophysical Journal, 193, 443.
Thorne, K. S. (1969a). Astrophysical Journal, 158, 1.
Thorne, K. S. (1969b). Astrophysical Journal, 158, 997.
Thorne, K. S. (1972). In Magic Without Magic: John Archibald Wheeler.AppendixA,
p. 243. W. H. Freeman: San Francisco.
Thorne, K. S. (1977). In de Sabbata and Weber (1977), p. 1.
Thorne, K. S. (1978). In Theoretical Principles in Astrophysics and Relativity, eds. N. R.
Lebovitz, W. H. Reid and P. 0. Vandervoort, p. 149. University of Chicago Press:
Chicago.
Thorne, K. S. (1980a). Reviews of Modern Physics, 52, 285.
Thorne, K. S. (1980b). Reviews of Modern Physics, 52, 299.
Thorne, K. S. and Braginsky, V. B. ( 1976). Astrophysical Journal (Letters), 204, L l .

Thorne, K . S . and Campolattaro, A . ( 1967). Astrophysical Journal, 149, 59 1 ; and 152,
673.
Thorne, K. S . , Caves, C. M ., Sandberg, V . D., Zimmermann, M. and Drever, R . W . P.
( 1979). In Smarr ( 1979). p . 49.
Thorne, K. S . , Drever, R . W . P., Caves, C. M., Zimmermann, M. and Sandberg , V . D.
( 1978). Physical Review Letters, 40, 667.
Thorne, K. S. and Giirsel, T . ( 1983). Monthly Notices of the Royal Astronomical Society,
205, 809.
Thorne, K. S. and Kovacs, S. J. ( 1975). Astrophysical Journal, 200, 245.
Thorne, K . S . , Price, R . H . and Macdonald, D . M . , eds. ( 1986). Black Holes: The
Membrane Paradigm. Yale University Press : New H aven, Connecticut .
Tipler, F . J . ( 1980). Physical Review D , 22, 2929.
Tourrenc, P . ( 1978). General Relativity and Gravitation, 9, 123 and 14 1 .
Tourrenc, P . and Crossiord, J.-L. (1974). Nuovo Cimento , 19B, 105.
Traschen, J . , Tu rok, N. and Brandenberger, R. ( 1986). Physical Review D, 34, 9 19.
Tsvetkov, V . P . ( 1 984). Soviet Astronomy - AJ, 28, 394.
Turner, M. and Wagoner, R . V. ( 1979) . In Smarr ( 1979), p. 383.
Turner, M. and Will, C. M. ( 1 978). Astrophysical Journal, 220, 1 107 .
Turok, N . and Brandenberger, R . H . ( 1986). Physical Review D, 33, 2 175.
Tyson, J . A . and Giffard , R. P . ( 1978). Annual Reviews of Astronomy and Astrophysics,
16, 52 1 .
Unru h , W . G . ( 1982). Unpublished work reported at NATO Advanced Study Institute
on Gravitational Radiation , Les Houches, France, June 1982.
Vachaspati, T. and Vilenkin , A. ( 1985). Physical Review D, 31, 3052 .
van den Heuvel, E. P . J . ( 1 984). In Reynolds and Stinebring, ed. ( 1984).
Vilenkin, A . ( 198 1a). Physical Review D, 24, 2082.
Vilenkin, A . (198 l b) . Physics Letters, 107B, 47.
Vinet, J. Y. ( 1985). Anna/es de Physique, 10, 253 .
Vinet, J . Y . ( 1986). Journal de Physique, 47, 639.
Vishniac , E. T. ( 1982). Astrophysical Journal, 257, 456.
Wagoner, R. V. ( 1977). In Gravitazione Sperimentale, ed . B. Bertotti . Accademia
Nazionale <lei Lincei : Rome.
Wagoner, R . V . ( 1984) . Astrophysical Journal, 278, 345.
Wagoner, R. V. and Will , C. M. ( 1976). Astrophysical Journal, 210, 764.
Wahlquist, H. D. ( 1987). General Relativity and Gravitation (in press).
Wahlquist, H . D . , Anderson , J. D . , Estabrook, F. B. and Thorne, K. S. ( 1 977). In
Gravitazione Sperimentale, ed. B . Bertotti , p. 335. Accademia Nazionale <lei Lincei :
Rome.
Wainstein , L . A . and Zubakov, V. D. ( 1962). Extraction of Signals from Noise, Prentice-
H all : London.
Wald, R . M. ( 1 973). Journal of Mathematical Physics, 14, 1453.
Walgate, R . ( 1983). Nature, 305, 665.
Walker, M. ( 1983). In Relativistic Astrophysics and Cosmology, eds. X . Fustero and
E. Verdaguer. World Scientific : Singapore.
Walker, M . and Will, C. M . ( 1980). Astrophysical Journal (Letters) , 242, L 129.
Walls, D . F. ( 1983 ). Nature, 306, 14 1 .
Weber, J . ( 1959) . Reviews of Modern Physics, 31, 68 1 (see especially Section V).
Weber, J . ( 1 960) . Physical Review, 1 17, 306.
Weber, J. ( 1 967). Physical Review Letters, 18, 498 .
Weber, J . ( 1 969). Physical Review Letters, 22, 1302.
458 K. S. Thorne
Weber, J. and Wheeler, J. A. ( 1957). Reviews of Modern Physics, 29, 509.

Weyl, H . ( 1922). Space-Time-Matter. Methuen : London .
Weisberg, J. M . and Taylor, J. H . ( 1984). Physical Review Letters, 52, 1348.
Weiss, R. ( 1972). Quarterly Progress Report of the Research Laboratory of Eleccronics of
the Massachusetts Institute of Technology, 105, 54.
Weiss, R. ( 1978). In Sources of Gravitational Radiation, ed . L. Smarr, p. 7. Cambridge
University Press.
Weiss, R. ( 1 979). In Smarr ( 1979), p. 7.
Weiss, R., Bender, P. L., Misner, C. W. and Pound , R. V . ( 1976). Report of the Sub
Panel on Relativity and Gravitacion, Management and Operations Working Group for
Shuttle Astronomy. NASA : Washington, DC.
Westpfahl , K. ( 1985). Fortschritte der Physik, 33, 4 17.
Wheeler, J. A . ( 1955). Physical Review, 97, 5 1 1 .
Wheeler, J. A . ( 1962). Geometrodynamics. Academic Press : New York .
Wheeler, J. A . ( 1964). In Relativity, Groups, and Topology, eds. C. DeWitt and
B. DeWitt. Gordon and Breach : New York .
Whitcomb, S . and Saulson, P. ( 1984). Unpublished research at Caltech and MIT ; a few
of the results are given in Fig. E.3 of Drever, Weiss et al. ( 1985).
Wiener, N. ( 1949). The Extrapolation, Interpolation and Smoothing of Stationary Time
Series with Engineering Applications. Wiley : New York .
Will , C. M . ( 1986). Canadian Journal of Physics, 64, 140.
Wilson, J. R. ( 1974). Physical Review Letters, 32, 849.
Winkler, W. ( 1977). In Gravitazione Sperimentale, ed . B. Bertotti. Academia Nazionale
<lei Lincei : Rome.
Winkler, W., Maischberger, K., Rudiger, A . , Schilling, R ., Schnupp, L. and
Shoemaker, D. ( 1986). In Ruffini , ed . (1986).
Witten, E. ( 1984) . Physical Review D, 30, 272.
Wood, K. S . , Michelson, P. F . , Boynton, P . , Yearian, M . R., Gursky, H . , Fried man , H .
and Dieter, J . ( 1986). A Proposal to NASA for an X-Ray Large Array (X LA) for the
NASA Space Station. Stanford : Palo Alto, California.
Wood, L., Zimmermann , G., Nukolls, J. and Chapline, G. ( 197 1). Bulletin of the
American Physical Society, 16, 609.
Woosley, S. E. and Weaver, T. A . ( 1986). Annual Reviews of Astronomy and
Astrophysics, 24, 205 .
Wu , L.-A . , Kimble, H . J . , Hall, J. L. and Wu , H . (1986). Physical Review Letters, 57,
2520.
Yamamoto, Y. and Haus, H. A. ( 1986). Reviews of Modern Physics, 58, 100 1 .
York, J . W . ( 1983). I n Deruelle and Piran ( 1983), p . 175.
Yurtsever, U. ( 1986). Abstracts of Contributed Papers, 1 1 th International Conference on
General Relativity and Gravitation, Stockholm, Sweden , 6-12 July , 1986 , p. 287.
University of Stockholm : Stockholm.
Yurtsever, U. ( 1987a). Physical Review D, submitted.
Yurtsever, U. ( 1987b). Paper in preparation.
Zel'dovich, Ya. B. ( 1980). Monthly Notices of the Royal Astronomical Society, 1 92, 663 .
Zel'dovich, Ya. B . and Novikov, I . D. ( 1983). Relativistic Astrophysics Vol. 2 . The
Structure and Evolution of the Universe. University of Chicago Pres s : Chicago.
Zhang, X.-H. ( 1986). Physical Review D, 34, 99 1 .
Zerilli , F . J . ( 1970). Physical Review D, 2, 2 14 1.
Zimmermann, M . ( 1978). Nature, 271, 524.
Zimmermann, M . ( 1980). Physical Review D, 2 1, 89 1.
10
The emergence of structure in the universe:
galaxy formation and 'dark matter'
M A RT I N J . R E E S
10. 1 Introduction
The first relativistic solutions for a homogeneous expanding universe were
found by Friedmann ( 1 922) before Hubble ( 1929) discovered the recession of
the nebulae. Hubble's work , which showed that the universe did not
resemble Einstein's ( 1 9 17) static model , stimulated further studies of
relativistic cosmology by Lemaitre, Tolman and others. But there was then
and remained for several decades - a severe mismatch between the relative
sophistication of the theory and the sparseness of the relevant data.
Hubble's work suggested that galaxies would have been crowded together
in the past , and emerged from some kind of 'beginning'. But he had no direct
evidence for cosmic evolution. Indeed the steady state theory, proposed in
1948 as a tenable alternative to. the 'big bang', envisaged continuous
creation of new matter and new galaxies, so that despite the expansion the
overall cosmic scene never changed.
We would not expect to discern any cosmic evolutionary trend unless we
can probe out to substantial redshifts. This entails studying objects billions
of light years away with redshifts z ::::: 1 ; although a programme to measure
the cosmic deceleration was pursued for many years with the 200-inch
Palomar telescope, the results were inconclusive, partly because normal
galaxies are not luminous enough to be detectable at sufficiently large
redshifts. It was Ryle and his colleagu�s, in the late 1950s, who found the first
evidence that our entire universe was evolving. His radio telescope could
pick up emission from some active galaxies (the ones thought to harbour
massive black holes) even when these were too far away to be observed
optically. One cannot determine a redshift or distance of these sources from
radio measurements alone, but Ryle assumed that , statistically at least, the
ones appearing faint were more distant than those appearing intense. He
460 M . J . Rees
counted the numbers with various apparent intensities, and found that there
were too many apparently faint ones (in other words, those at large
distances) compared with brighter and closer ones (Ryle, 1 958). This was
discomforting to the 'steady statesmen' , but compatible with an evolving
universe if galaxies were more prone to violent outbursts in the remote past,
when they were young. The subsequent discovery by optical astronomers of
many hundreds of active galaxies at very large redshifts (quasars) has borne
out this trend ; but these objects, and their evolution, are still too poorly
understood to be used for determining the geometry and deceleration of the
um verse.
The clinching ·evidence for a 'big bang' came when Penzias and Wilson
( 1965) detected the cosmic microwave background radiation. This radiation
(whose thermal spectrum was quickly established) implied that intergalactic
space was not completely cold, but at a temperature of 2.7 K. The
corresponding photon density is "' 4 x 108 m 3 , implying that there are
-
,..., 109 photons for every baryon. This discovery quickly led to a general
acceptance of the so-called 'hot big bang' cosmology - a shift in the
consensus among cosmologists as sudden and drastic as the shift of
geophysical opinion in favour of continental drift that took place
contemporaneously. There seemed no plausible way of accounting for the
microwave background radiation except on the hypothesis that it was a relic
of an epoch when the entire universe was hot, dense and opaque. Moreover,
the high intrinsic isotropy of this radiation - better than one part in 10 4 -
meant that the Robertson-Walker metric was a better approximation to the
real universe than the theorists of the 1930s would have dared to hope.
In the late 1960s some theorists were emboldened to carry out a series of
now-classic investigations of the early stages of a Friedmann universe
composed of matter and radiation . Insofar as the present 'mix' of matter and
thermal radiation was known, cosmologists could infer the appropriate
equation of state at earlier times, and deduce the universe's 'thermal
history'. Powerful corroborative support for the hot big bang came when the
composition of material emerging from the 'fireball' was calculated , and
found to be 25 per cent helium and 75 per cent hydrogen (Hoyle and Tayler,
1964; Peebles, 1966 ; Wagoner, Fowler and Hoyle, 1 967). This was specially
gratifying because the theory of stellar nucleosynthesis, which worked so
well for carbon, iron, etc., was hard-pressed to explain why there was so
much helium, and why its abund.ance was so uniform. Attributing most of
the observed helium to the big bang therefore solved a long-standing
problem in nucleogenesis, and bolstered cosmologists' confidence in
Galaxy formation and dark matter 461
extrapolating right back to the first few seconds of the universe's history, and
assuming that the laws of microphysics were the same then as now. The
applicability of the Friedmann equations at early times c� uld not of course
be taken for granted ; the universe is certainly not completely homogeneous
now, and could in principle have been more irregular, and more anisotropic,
in the past. But the measured isotropy, together with the singularity
theorems, implied that there must be some singularities in the universe's past
(Hawking and Ellis, 1968), and the observed helium abundance (sensitive to
the expansion rate when the temperature was - 101 ° K) constrains any
possible early anisotropies.
More detailed calculations, combined with better observations of
background radiation and of element abundances, have strengthened the
consensus that the hot big bang model is basically valid . Several discrepant
results could have emerged during the last 20 years, but did not : for instance,
the standard Friedmann hot big bang would need drastic modification if any
object were found to have a zero helium/hydrogen ratio, if any species of
neutrino were found experimentally to have a mass in the range 1 keV-
1 Me V , or if there were a glaring discrepancy between the ages of the oldest
stars and the largest plausible timescale in such models. The hot big bang is
not yet fi rm dogma. Conceivably, our satisfaction will prove as transitory as
that of a Ptolemaic astronomer who successfully fits a new epicycle. But the
hot big bang model certainly seems more plau sible than any equally specific
alternative - most cosmologists would make the stronger claim that it has a
more than 50 per cent chance of being essentially correct.
The main stages in the evolution of a standard hot big bang universe are
depicted schematically in Fig. 10. 1 . Uncertainties about the relevant physics
impede our confidence in discussing the extensive span of logarithmic time
10 - 43_ 10 - 4 s when thermal energies exceed 100 MeV . When t ;(; 10 - 4 s we
can consistently utilise physics which is 'well known'; and, so long as the
universe remains almost homogeneous, the evolution is straightforwardly
calculable. However, at some stage small initial perturbations must have
evolved into gravitationally bound systems (protogalaxies?); even though
the controlling physics may then be Newtonian gravity and gas dynamics,
the onset of non-linearity induces challengingly complex behaviou r.
The progress of the last 20 years in delineating the big bang model has
therefore brought two sets of questions into sharper focus.
(i) How did the universe behave during the very earliest phases when the
physics is uncertain? Such fundamental properties as its scale, overall
462 M. J . Rees
homogeneity and matter content could have been imprinted during

these initial stages.
(ii) How did the dominant present-day structures in our unive�se - galaxies
and clusters - evolve from amorphous beginnings?
The first set of questions is addressed elsewhere in this volume. I shall
focus on the second topic, and outline some current ideas on the inter
related issues of galaxy formation , dark matter, the density of the universe ,
and the nature of the initial fluctuations.
Observational cosmology reveals that there may be just three (or perhaps
four) important constants which characterise the universe's overall
properties.
(i) The Robertson-Walker curvature radius. The fact that the universe is
still expanding , after 1060tp1an ck , with a density within an order of
magnitude of the 'critical' density , implies that the initial curvature
Fig . 10. 1 . A schematic depiction of the history of a standard Friedmann hot

big bang model , from the Planck time to the p resen t .
Relevant tnow
--- Most distant quasar
astrophysics
not well known !
--- Last scattering of
· microwave background
Relevant physics.
well known
1 s }>--- Helium synthesis
--- Quark � nucleon transition

--- Salam-Weinberg phase transition
I
I
I
I
I
Relevant physics ' Desert '
speculative I
I
I
I
I
Relevant physics
{
l 1 0-35 s
)--- Inflation
I
Baryosyn thesis (- nb/ny)
(� ' flatness '
+ fluctuations?)
very speculative
(or unknown)
1 o-43 s tPlanck Q uantum gravity
radius at t P l anck (or at the end of an 'inflationary' phase) is at least "" 10 3 0

times larger than the horizon scale at that epoch. The precise value of
this curvature radius depends on the density parameter n.
(ii) The photon-to-baryon ratio. This ratio is a measure of the entropy per
baryon . Grand unified theories (GUTs) suggest that it can be explained
in terms of baryon non-conservation processes.
(iii) The fluctuation amplitude. The prime mystery is perhaps why the large
scale universe is so homogeneous. However, some fluctuations are
essential in order to trigger galaxy formation. If these metric
fluctuations have a scale-independent random-phase character, then
the amplitude is pinned down to be 10 - 4- 10 - 5 , a number which
theorists may hope to calculate; however, if the fluctuations have a
more general character, few constraints can be set.
(iv) A possible fourth constant appears if the universe contains a
dynamically-significant amount of non-baryonic matter. The fraction
of mass-energy in baryons rather than more exotic species is then
determined by physical processes in the high-density phases of the big
bang .
Insofar as the aim of fundamental physics is to erode the number of
independent underivable constants, it is gratifying that there is a serious
chance of calculating the above numbers in terms of microphysical
parameters. And then , given that the universe is sufficiently close to
'flatness', but nevertheless contains fluctuations (i.e. there are small
amplitude 'ripples' on the initial hypersurface), gravitationally bound
structures are fated eventually to emerge. Such structures - the galaxies -
then offer an environment where stars, some with planetary systems, form
and evolve.
10.2 The constituents of the universe: 'dark' matter and 'luminous'

matter
To the cosmologist, entire galaxies are just 'markers' or test particles
scattered through space which indicate how the material content of the
universe is distributed and how it is, moving. But it is these 'small-scale'
deviations from homogeneity, groups of galaxies and their constituent
members, which constitute the main subject matter of astronomy. And it is a
prime goal of our subject to understand how the universe has evolved from
an initial dense fireball 101 0 years ago to its present state, where galaxies
""
dominate the large-scale cosmic scene.

An impediment to understanding galaxies and their spatial distribution is
464 M. J . Rees
that the stars and gas we observe may be little more than a tracer for the
material that is dynamically dominant. The evidence for 'dark matter' dates
back more than fifty years, but has firmed up since the classic papers of
Einasto, Kaasik and Saar ( 1974) and Ostriker, Peebles and Yahil (1 974) and
is now quite compelling (for a recent review, see the proceedings of IAU
Symposium 1 17 (Knapp and Kormendy, 1986)); what the dark matter
consists of is , however, still a mystery.
The masses inferred from relative motions of galaxies in apparently-bound
groups and clusters exceed by a factor 10 those inferred from the internal
-
dynamics of the luminous parts of galaxies. This apparent discrepancy could

be resolved if galaxies were embedded in extensive dark 'halos'. This
hypothesis can be checked in some edge-on disc galaxies, where emission
from gas can be observed at radii far exceeding the extent of the conspicuous
stellar disc. The mass of this gas is itself negligible, but rotation velocities
derived from its spectral lines do not fall off as r - 1 12 , as would be expected if
the gas were orbiting a mass distribution concentrated at much smaller
radii . Instead , the velocity remains almost constant, implying M( < r) oc r,
out to 80 kpc in some cases. Direct lower limits on the mass to light ratio
-
M/L can be obtained for the matter in the outlying parts of galaxies with
measured rotation curves, and for the haloes of edge-on galaxies. Values of
M/L exceeding 300 solar units are sometimes found (see Rubin (1986) for a
recent review).
Estimates of the masses of galaxy clusters come from the virial theorem - a
technique first applied to the Coma cluster by Zwicky ( 1 93 3). This method is
now complemented by X-ray studies of thermal emission from hot gas in
clusters, which probe the depth of the gravitational potential well. In well
studied clusters which appear to have reached virial equilibrium, M/L is
typically 200 solar units. This matter must mainly be in some unknown form
- neither ordinary stars nor the gas that emits X-rays. The data are
summarised in Figure 10.2.
All things considered, the existence of dark matter is quite unsurprising -
there are all too many forms it could take, and the aim of observers and
theorists must be to narrow down the range of options. These topics have
been discussed and reviewed at greater length elsewhere (e.g. Dekel, Einasto
and Rees, 1987, and references cited therein).
The present Hubble timescale t8 is still uncertain, so in quoting numerical
values I will follow a widespread convention and introduce a quantity h =
(3 x 10 1 7 s/t8). The experts advocate values of h in the range 0.5- 1 ; for a
detailed assessment, see Hodge ( 198 1) or Rowan-Robinson ( 1985). The
Fig . 10.2. (a) The apparent increase with scale of the mass-to-light ratio .
This increase is due to the two distinct trends : (i) ordinary stars and gas are a
decreasing proportion of the mass of the larger systems; and (ii) in the larger
systems, even the 'ordinary' (star + gas) components have a larger M/L,
because they consist primarily of elliptical galaxies with few young stars, and
contain much hot gas revealed only by its X-ray emission. (b) The same data
re-plotted , with effect (ii) subtracted out. One finds that the physically more
fundamental ratio of 'ordinary' matter to 'dark' matter is independent of scale
in all virialized systems larger than galaxies, and has a value consistent with
!1 = 0. 1--0.2. The situation is less clear on still larger scales (superclusters)
because the dynamics are uncertain and virial equilibrium does not prevail.
This figu re is adapted from Faber ( 1984) and Blumenthal et al. ( 1984) ; the
latter paper gives fuller details of the data on which it is based . (The masses of
dwarf spheroidal galaxies are uncertain and controversial ; the diagrams
show these systems plotted twice, depending on whether or not they contain
dark matter.)
� cores
(a)
Small E groups �
M
L�e
2 way cluster
...;i
�
�
Dynamical mass m
I
S2
tll)
�
Dw arf spheroidals
Small spiral groups
arrnfurn Luminous mass
2 (b )
M w ay
Large
]
cluster cores
Dynamical mass I
� trn �
�
tll)
S2
'1°
Dw arf spheroidals Small spiral groups
0 mmfum Luminous mass
6 8 10 12 14 16
log M
466 M. J . Rees
= 3,
mean baryon density is then nb 1 l !lbh 2 m - where nb is defined as the
fraction of the critical density Pcrit (JnGt�) - 1 that is in baryonic form. An
=
important number which is perhaps of fundamental significance is the ratio

of the baryon density to the density n y of photons in the microwave
background, apparently a black body with T -::::::, 2.7 K. This is then
y - 1 = nb/ny - 3 x 10 - s
( T )-3 (Qbh2),
2 .7 K
(2 . 1)
fl' being a measure of the entropy per baryon ; it is a number which GUT
models must attempt to explain .
A direct lower limit on nb of order 10 - 2 can be set from the observed
mean density of baryons in conspicuously 'luminous' form (visible galaxies,
and intergalactic gas revealed by its X-ray emission), implying a baryon-to
photon ratio of not less than about 3 x 10 - 1 0 h 2 .
There is no firm evidence for any antimatter in the universe (apart from a
small fraction of antiparticles in cosmic rays, which could have been
produced in high-energy collisions). Strong constraints on the presence of
antimatter in and around our Galaxy are set by the measured limits to the X
ray background . Nevertheless, if one is strictly agnostic and free from
theoretical preconceptions, one can certainly envisa,ge that the universe
might possess matter-antimatter symmetry (i .e. that the overall net baryon
number, and ( nb/n y ) , are zero) provided that the scale of the regions of each
'sign' is at least as large as a cluster of galaxies.
The inferred dark matter in the halos of individual galaxies and in clusters
of galaxies apparently contributes a fraction n = 0. 1 -0.2 of the critical
cosmological density. Its smoother and less clumped distribution suggests
that it underwent less dissipation during the processes of galaxy formation
than the luminous stars and gas. (In a final section I shall address the
separate question of whether the data are compatible with there being still
more dark matter; enough, in particular, to provide the entire critical
density (Q = 1).)
Three strands of evidence could eventually pin down what the dark
matter in halos and clusters of galaxies really is.
(i) Particle physics. When our theories of high-energy physics become less
speculative, and we can calculate how many particles of each species
(with known mass) should have survived as relics of the big bang, it may
turn out that at least one specific species with non-zero rest mass is
predicted , on the basis of standard cosmology, to contribute
significantly to n.
(ii) Cosmogonic models. Evidence on the clustering scales of galaxies, the

density profiles in halos, etc. , can be compared with the outcome of
simulations of gravitational aggregation. The character of the
clustering depends , in particular, on whether the dark matter is 'cold'
(in the sense that its thermal velocities are unimportant on relevant
scales) or 'hot'.
(iii) Direct detection. The individual entities that ou r galactic halo is
composed of may reveal themselves by astronomical observations, or
by direct Earth-based experiments.
1 0 .2 .1 Baryonic dark matter

The baryons observed in galaxies and gas, contributing nb � 0.0 1, could in
principle be a dynamically unimportant fraction of all baryons in the
universe. Most of the initial baryons might have condensed into a
population of stars that were either pregalactic, or else formed during the
initial collapse phase of protogalaxies : could these stars, or their remnants,
contribute to the unseen mass? Ideally one would like to be able to calculate
what happened when the first g ravitationally bound clouds condensed from
primordial material - to decide whether they form one (or a few) very
massive objects (VM Os) , or whether fragmentation proceeds efficiently
down to low-mass stars. Our poor understanding of what determines the
initial mass function (IM F) of stars forming now (in, for instance, the Orion
nebula) gives us little confidence that we can calculate the nature of stars
born in an envi ronment very different from our present-day galaxy.
Constraints imposed by background light, limits to the heavy element
abundance and absence of dynamical effects in the halo (see, for example,
Carr, Bond and Arnett , 1984 ; Rees, 1986) imply that the individual masses
of halo objects are either less than 0. 1 M 0 or else in the range 102- 106 M 0 .
These constraints are summarised in Fig . 10.3 and its caption.
Two mass-scales that emerge naturally in cosmogony are 106 M 0 and
10 - 2 M 0 . The Jeans mass in the baryonic component of the universe after
recombination is of order 106 M 0 ; a similar mass appears in a different, but
equally relevant, context, as the Jeans mass in 10 4 K gas that is in pressure
balance with hotter gas in a collapsing protogalaxy (Fall and Rees, 1985). If
a 106 M 0 cloud collapses isothermally the Jeans mass falls, however, to
10 - 2 M 0 before the onset of opacity effects; this is the case even in a
primordial cloud containing no heavy elements. The IMF therefore depends
crucially on the complexities of the fragmentation process. Does
468 M. J . Rees
fragmentation of the first clouds proceed right down to the 'opacity limited'
Jeans mass? Or, is fragmentation impeded by collisions (and coalescence) or
protostars, or by tidal effects? I do not believe we can yet answer these
questions with any confidence; it is therefore worth considering both
'Jupiters' and 'VMOs' seriously, in the hope that observations can offer
some firmer clues than theory.
One way of discriminating between the 'Jupiter' and 'VMO' options is by
searching for evidence of gravitational lensing. The probability of seeing
lensing due to an object in our own halo is only of order 10 - 6 (Refsdal , 1964 ;
Paczynski, 1986b) ; but the cross-section for effective lensing is proportional
Fig. 10.3. This diagram, from Carr, Bond and Arnett ( 1984) shows various
constraints on the fraction n* of the critical cosmological density that can be
contributed by first-generation stars (or their remnants) in different mass
ranges . The objects are assumed to form at a redshift Zr . There are dynamical
constraints on the number of �. 106 M 0 black holes in galactic halos, because
such massive bodies would have sedimented inwards via dynamical drag on
the ordinary stars. The requirement not to overproduce heavy elements
constrains the number of remnants of ordinary heavy stars which end their
lives by exploding as supernovae. Stars in the mass range 0. 1 - 1 M 0 would
still be shining after - 1010 yr, producing too much background light. The
possible options for dark matter are 'Jupiters' of below 0. 1 M 0 , or the
remnants of very massive objects (V MOs) in the mass range from a few
hundred to 106 M 0 , which could have formed at a large zr. (These latter
objects do not necessarily eject any material processed beyond helium, and
leave black hole remnants.) One way of deciding between these two
possibilities is by seeking evidence for gravitational lensing, as discussed in
the text.
1 0- 1
1 0 -2
n.
1 0 -3
'--�-'-�---'-�---'L--�-'--�____L_�--L��L-�...L__J
I I I
1 0-4
I
4 8 15 Mc
1 0 -s
1 0- 1 10 1 02 1 03 1 04 1 05 1 06 1 07
M(M 0)
to distance, so there is, perhaps surprisingly, much more chance of detecting

objects in the halos of galaxies (or in clusters) half way out to the Hubble
radius. The probability that a compact source at redshift z ;;;::; 1 is significantly
lensed by objects along the line-of-sight is of order nl , independent of the
individual lens mass involved ( Refsdal , 1970; Press and Gunn, 1973); for a
source at redshift zs < 1 the probability is "" z;n1• The angular separation, 81 ,
of the lens images is, however, a diagnostic of the masses :
81 � 2 x 10 - 6 (M1 /M 0 ) 1'2 max [zs- 1 12 , 1] . (2.2)
For M1 ;;;::; 105 M 0 , very long baseline radio interferometers provide
adequate resolution. Characteristic image shapes were discussed by Press
and Gunn ( 1973), and Blandford and Jaroszynski ( 1 98 1). We can probably
exclude n 1 in 106 M 0 objects.
=
For M ;S 0. 1 M G ('Jupiters') the angular scale is less than ( 10 - 6)". This

cannot be directly resolved by any technique, until optical interferometers
are deployed in space. There is nevertheless a genuine prospect of detecting
lensing of this kind because of the variability that would ensue if the lens
were to move transversely (Gott, 198 1 ; Young, 1 98 1) : an object at the
Hubble distance moving at ;;;::; 10 2 km s - 1 takes only a few years to traverse
an angle ( 10 - 6)" . The image structu re and time variation are more
complicated if the line-of-sight passes through , for example, a galactic halo,
thereby encountering an above-average column density of dark matter.
Several objects may then contribute to the imaging (Chang and Refsdal,
1979; Young , 198 1 ; Paczynski , 1986a) yielding a 'frosted glass' effect.
'Minilensing' requi res a source whose intrinsic angular size is smaller than
the value of 81 given by (2.2) . Possibilities include distant supernovae
(Efstathiou, Schneider and Wagoner, 1986) or the region that emits the
optical continuum of quasars (but not the spectral lines, because the latter
come from a more extended region) . If there were a firm observational limit
to the scatter in the equivalent widths of the lines from quasar to quasar (i.e.
in the line-continuum ratio) this would constrain the value of n contributed
by small compact objects. Canizares ( 1 982) uses this argument to rule out
!1 = 1 in any objects with mass greater than 10 - 2 M G ; n � O. l cannot yet be
excluded, but the method holds prospects of improvement.
1 0 .2 .2 Constraints on Qb from primordial nucleosynthesis

If we accept the concept of light element synthesis in the big bang (when kT
is in the range 0. 1- 1 Me V) then an important constraint on the total
baryonic density nb comes from deuterium; this is an intermediate product
470 M. J. Rees
in helium synthesis, the amount that survives being a steeply decreasing

function of nb. According to Yang et al. ( 1984) only if Qbh 2 � 0.04 can
enough deuterium be produced in a standard hot big bang to provide the
observed abundance ; and Qbh2 ;<:: 0.0 1 is required in order not to
overproduce deuterium or 3 He. Also, weak constraints consistent with the
other requirements are provided by 7Li.
Because the relevant parameter in primordial nucleosynthesis is nb/ny oc
Qbh2, more precise comparisons of models with observation must await a
firmer value of the Hubble constant . If h = 1 (corresponding to a Hubble
time of 10 1 0 years) then the simplest inference would be that most of the
dynamically detected dark matter - both in the halos of individual galaxies
and in clusters and groups - was non-baryonic ; but if h -1- (corresponding
=
to a Hubble time of 2 x 101 0 years) the lower limit to Qb set by the

requirement not to overproduce D + 3 He implies that some dark matter -
maybe that in halos, if not in intergalactic space - must be baryonic, though
only enough to contribute nb ::::::: 0. 1 . The strength of this limit (which of
course precludes nb 1 for any reasonable Hubble constant) depends on
=
how compelling the case for 'standard' big bang nucleosynthesis actually is.
Even in the context qf this theory, Applegate, Hogan and Scherrer (1986)
have recently · pointed out that small scale inhomogeneities in the baryon
photon ratio, such as might arise at the quark-hadron transition, could
modify the resultant abundances : if the scale of the inhomogeneities is
smaller than the neutron diffusion scale at t ::::::: 1 s, then the neutron-proton
ratio can vary from place to place in a manner that would be impossible in
any homogeneous model , with the result that the 4He abundance can be
lower than in the standard model , and the observed deuterium could be
primordial even if nb = 1 .
1 0 . 2 .3 Non-baryonic dark matter

If neutrinos have negligible rest mass, the present density expected for relic
neutrinos from the big bang is n v 1 . 1 x 108(T/2.7 K) 3 m - 3 for each two
=
component species. This conclusion holds for non-zero masses, provided

that mvc 2 is far below the thermal energy (5 MeV) at which neutrinos
decoupled from other species and that the neutrinos are stable for the
Hubble time. Comparison with the baryon density shows that neutrinos
outnumber baryons by such a big factor ( "' Y') that they can be dynamically
dominant over baryons even if their masses are only a few electron volts. In
fact, a si�gle species of neutrino would yield a contribution to n of nv =
0.densi
0 1 ty. so if 0. 5 , only 25 eV is sufficient to provide the critical
h - 3 (m.Jev ; h=
The entire range 100h2 eV-3 GeV is i

model (Gunn et al. , 1978). (For my <: 3 GeV, the rest mass term in then compati b le wi t h the hot bi g bang
Boltzmann factor woul d ki l l off most of the neutrinos
the number surviving would be less than nb ). If any species of neutrino were before they decoupl e d;
discovered
cannot to have
extrapolate a mass i n this excluded range, it woul d show that one
the hot bi
the photons must have been generated at later times.g bang back to k T <: 5 MeV, and that most of
More
Szalay than 15 years ago, Cowsik and McLelland (1973) and Marx and
(1972) conjectured that neutrinos could
galactic halos and clusters. At that time the suggestion was not followed up provide the ' u nseen' mass i n
very
about extensi
the v ely;
possibility but ofin the
non-zero 1980s physi
neutri n oc i s ts
masses.became A more
change open-mi
of theoretin ded
c al
atti1980)tude,stimcoupled with experi m ental clai m s that my � 36 eV (Lyubim ov et al. ,
whi c h ulated
neutri n o astrophysi
clusteri n g andc istsdito
ffuexplore
sion scenarios
play a key for
role galaxy
(see formation
Section 10. 4in).
More recently, other kinds of non-baryonic matter, such as supersymmetric
particles and axions, have also been considered.
Provided
speci e s of that
el e we
mentary knowparti thec mass
le, we and
can anni
i n prih i
n l ation
ci p le cross-secti
cal c ul a te o n
how for any
many
survive from
topartinc. leProgress the bi
ing bang, and
experimental the resultant
parti c le contri
physi c sb ution
may each
thereforespeci e s makes
reveal a
whi c h must
big bang theory entirely.dynami contribute si g
Indeed,calonceni fi c antly
we to
admi n, unless
t the we
possi abandon
b i l i t y that the hot
non
baryoni c matter may be another fundamentally important,number the ratioasnnon-baryonic/nb
important for
i n
the universe becomes
cosmology as n b /n .
y
Non-baryonic matter, dynami(see

if process callySectidomionn10.ant,4). Insofarhas an asimportant we come
influence on the cosmogonic
galaxy formation, we may therefore acquire circumstanti a l
to understand whether suchone matter exists and whatsuchformpartiit cmilesght(eventake.in
evidence about that maythebe mass able to detect
However, it turnsif out theyng contribute ofpartiourclesownwouldgalactihave c halo.cross
the laboratory) interacti cmmassivfore interactions ( <: 1 GeV) with nucleons. Such particles,
Weaklyof order 6 2
10- off nucleons in the Sun, would lose energy via the
sectionsng elastically 3 -
scatteri
recoil, and could thereby become trapped (Stei et al. , 1978; Spergel
gmanby ordinary
and Press, 1985). Despite being vastly outnumbered nuclei, they
472 M. J . Rees
could then contribute to energy transport in the solar core, because their
mean free path is so long ; the central temperature would then be slightly
lower, with resultant observational consequences such as an alteration in
the frequency of some modes of solar oscillation, and a reduction in the 8 B
neutrino flux . Over the lifetime of the Sun , an isothermal core of 'inos' could
build up a mass of 10 - 1 2 M 0 if annihilations did not occur. Annihilations
would restrict this buildup, unless the cross-section for annihilation is far
below that for scattering , or unless the big bang produces an excess of inos
over anti-inos (as it does for baryons). However, even if annihilations
prevent a dense enough core building up to affect the Sun's structure, energetic
neutrinos from these annihilations may reveal their presence in the
underwater detectors developed to search for proton decay. Already, scalar
or Dirac neutrinos with mass exceeding 6 GeV can be excluded , and
analogous limits come from considering annihilations in the Earth rather
than the Sun (Silk et al., 1985).
If our galactic halo were comprised of massive weakly interacting
particles, then their density near the Earth would be "" 105mGev-1 m 3 , and
their typical velocities "" 300 km s - 1 • There is a genuine prospect of
-
detecting this background in the laboratory. Whenever one of these particles

collided with a nucleus, there would be a recoil , with associated energy in the
keV range. This energy deposition could be detected if it occurred in a very
cold solid (with low heat capacity oc T3). The detection could be achieved by
maintaining an array of superconducting grains just below the transition
temperature : the heat would then raise the temperature of one grain above
the critical value, thereby allowing magnetic flux to penetrate (Goodman
and Witten, 1985; Drukier, Freese and Spergel , 1986). Alternatively, the
phonons could be detected before thermalisation , or the thermal pulse could
be detected directly (Smith, 1986).
If the dark matter were in so-called 'invisible axions', there would be a
chance of detecting these also via conversion into photons on interaction
with matter or strong magnetic fields. The prospects here are perhaps rather
less promising , because the photons would be in a narrow energy b and ,
depending on the poorly known axion mass, so a broad range of photon
energies (spanning the microwave, millimetre and infrared bands) would
need to be searched (Sikivie, 1985).
Ingenious schemes for detecting a halo population of exotic particles
(reviewed by Smith, 1986) seem to me among the most worthwhile and
exciting high-risk experiments in physics or astrophysics - potentially as
important as those that led to the discovery of the cosmic microwave
background in the 1960s. A null result would surprise nobody ; on the other
hand , such experiments could reveal new supersymmetric particles (or
axions, as the case may be), as well as determining what 90 per cent of our
universe consists of. (Because the detection is sensitive to velocity, they
would even reveal the halo's velocity dispersion and rotation. The mean
velocity of halo particles relative to the detector would have an annual
variation , because of the Earth's motion around the Sun. Such an annual
modulation , with an amplitude of a few per cent and a peak in June , would
be an unambiguous signature discriminating against spurious background.)
1 0 .2 .4 Macroscopic 'cold dark-matter' candidates: nuggets of

strange matter
Witten (1984) conjectured that grains or nuggets of 'strange matter' ,
containing up, down , and strange quarks, may survive stably from the
quark-hadron transition at t = 10 - 4 (see Fig. 10. 1). Such objects, in some
sense intermediate between elementary particles and lumps of astrophysical
size , would count as non-baryonic matter in the context of nucleosynthesis.
Recent work (Applegate and Hogan , 1 98 5 ; Alcock and Fahri , 1985) suggests
that neutrino heating when kT -;:::; 1 MeV would destroy nuggets unless they
had a mass of planetary order; and it is unclear that any larger ones could
even form, because this would involve coordination over a scale larger than
the particle horizon at the relevant epoch. If they indeed had survived , these
nuggets might be detectable (De Rejula and Glashow, 1984) ; interesting
constraints on the number in various mass ranges could be set from the
results of monopole searches and proton decay experiments, from the
number of meteor showers, and from limits on the frequency of small-scale
seismic events.
1 0 .2 .5 The baryon-photon ratio

Ideas on baryon synthesis - if the GUTs on which they are based are borne
out by future developments - might allow us to test whether the Friedmann
36
models apply back at temperatures of around 10 1 5 Ge V, corresponding to
times ,...,, 10 - s. The observed baryon-to-photon ratio (eq . (2 . 1)), a measure
of the fractional excess of baryons over their antiparticles at early times, is
,...,, 10 - 9. (Were it much smaller than this, the universe would not be baryon
dominated when its age was of order a characteristic stellar lifetime.) The
value of the net baryon excess arising from out-of-equilibrium decay of X
and Y particles can be computed, given a specific GUT (Kolb and Wolfram,
1980) ; it involves a small parameter related to the CP-violation parameter in
474 M. J. Rees
weak interactions. This work is not yet on the same footing as the
calculations of primordial helium and deuterium ; it is perhaps at the same
level as nucleosynthesis was in the pioneering days of Gamow and Lemaitre .
But if it could be firmed up it would represent an extraordinary triumph.
The mixture of radiation and matter characterizing our universe would not
be ad hoc but would be a consequence of the simplest initial conditions. Also,
as well as vindicating a GUT, it would reassure us about extrapolating in
one bound, based on a Friedmann model, right back to the threshold of
classical cosmology, almost back to the Planck time. On a logarithmic scale,
this is a bigger extrapolation from the nucleosynthesis era than is involved in
going to that era from the present time. It would also place constraints on
dissipative processes arising from viscosity, phase transitions, black hole
evaporation , etc., which might occur as the universe cooled through the
'desert' between 10 1 5 and 100 Ge V. Al though these ideas are still
speculative, the 'prediction' of the photon-baryon ratio may turn out to
offer one of the few empirical tests of G UTs. (If baryon number were strictly
conserved and the universe actually possessed a conserved quantity of order
108 0 , then the concept of inflation would lose its appeal . The advent of
GUTs is therefore a prerequisite for the viability of such theories,
irrespective of the mechanism that drives the phase of exponential growth.)
If the baryon-photon ratio could be calculated , this would determine nb .
If Qb < 1 , then a strictly flat universe would require some non-baryonic
contribution. Of course , one may eventually have theoretical knowledge of
the rest masses of all other relevant particles; such information , in
conjunction with knowledge of nb/n y , would determine their contribution to
Qb also . Looked at from this point of view, it perhaps seems coincidental
that non-baryonic matter should dominate, but only by an order of
magnitude rather than a vastly larger factor.
10.3 Large-scale structure and isotropy

For galaxies to have formed , the early universe must have contained some
inhomogeneities (despite its overall Friedmannian character) ; otherwise its
baryon content would, after 10 1 0 years, still ju st comprise uniform neutral H
and He. Gas dynamical and dissipative effects must have been important in
the formation of individual galaxies (so that the characteristics of galaxies
are related only indirectly to the form of the primordial fluctuation
spectrum). But inhomogeneities on much larger scales are more likely to
have been induced primarily by gravitation . The clustering properties of
galaxies therefore offer more direct evidence about the initial fluctuation
amplitudes. Note, however, that we can use these to infer the total density
fluctuations only insofar as galaxies are a good tracer for the overall mass
distribution. Large-scale inhomogeneities would also induce 'peculiar
velocities' - deviations from the Hubble flow - in galaxies and even entire
clusters . The existing evidence on superclustering - though agreed by all to
be of primary importance - is still tentative and ambiguous.
A quite separate line of attack on large-scale structure is offered by the
background radiation. The microwave background isotropy (established
via relative rather than absolute measurements) is already known to be
amazingly precise : effects at the level of one part in 104 would be detectable
in a total cosmic background which is "' 102 weaker than the contribution
from the Earth itself. Recent reviews of the data have been given by
Partridge ( 1986), Wilkinson ( 1986) and Kaiser and Silk ( 1986). There are no
confirmed anisotropies apart from a 'dipole' anisotropy, of /J. T/T ';::::, 1 . 2 x
10 - 3 , indicating a motion of our Local Group of galaxies towards a
direction 45° from the Virgo cluster. The upper limits are of order 10 - 4 on
all angles from a few arc minutes up to 90° (quadrupole).
The microwave photons were last scattered at a 'cosmic photosphere'
whose redshift z* and thickness depend on the thermal history of the matter.
If the primordial plasma (re)combined as it cooled adiabatically below a
temperature of a few thousand degrees, and there was no significant later
reheating , then z* ';::::, 10 3 ; the effective thickness of the photosphere,
determined by the width /J.z* of the function (d/dz)e - r<z> , is then only
- 0. l z* . This is because the reduction in the free electron density (which
results in the 'fog lifting' and the microwave photons becoming free to travel
uninterruptedly) is rather sudden . If reheating caused the intergalactic
medium to 'fog up' again at a later epoch, with t > 1 at redshifts below 1000,
then z* would be smaller and /J.z* � z. Three cases are illustrated in Fig. 10.4.
Note that reheating at z < 10 could never generate t > 1, even if the universe
contained enough ionized intergalactic gas to give n = 1 .
A Friedmann model with !1 0 < 1 evolves very like a 'flat' model at
redshifts > !10 1 . Provided that z* > !10 1 , which is so for almost all relevant
models, the angle subtended by a given comoving region on the last
scattering shell is essentially independent of z* . A comoving length l
corresponds to an angle
(3 . 1)
A region of present length 1 Mpc subtends an·angle -!!l0 h arc minutes. A
region that came within the horizon at z* has present length l* � 2ctH x
(!l 0 z*) - 1 1 2 and subtends an angle ()* ';::::, (!l0 /z*) 1 12.
476 M. J. Rees
Another important scale, if Q0 < 1 , is the length corresponding to the

intrinsic curvature of the spacelike hypersurfaces in the Robertson-Walker
metric; this subtends an angle ec � !lo ( l - !lo ) - 1 1 2 .
The fluctuation amplitude on a given scale is conveniently characterised
{ }
by
value of ( ( bp/p�2 ) 11 2 wh�n scale I is first
s(l) = . . (3.2)
encompassed w1 thm particle honzon .
If the growth were unaffected by pressure, this spectrum leads to
( (bp/p)2 ) 1 '2 oc s(l) x 1 - 2 for all scales l within the horizon at a given time -
density contrasts, in the linear regime would grow oc R(t), and would
eventually condense out as virialised systems with gravitational binding
energy - sc 2 per unit mass. The quantity s is the 'curvature fluctuation',
which one would like to be able to calculate from first principles.
Any fluctuation straddling the cosmic photosphere whose comoving
length-scale exceeds l* , the horizon scale at z* , but is less than ctH would
generate a temperatu re anisotropy (� T/T) which is simply of order s, as was
first calculated by Sachs and Wolfe ( 1967). Similar fluctuations along the
line-of-sight with redshifts less than z* have smaller effects (Rees and
Sciama, 1968 ; Dyer, 1976). If z* � 1000, then all fluctuations subtending
angular scales '> (2!1�1 2 ) degrees exceed the horizon scale; even if z* � 10 (an
implausibly extreme lower limit, attained only if most of the critical density
Fig. 10.4. Optical depth • (z) (dot-dash lines) of the universe back to a
redshift z, and 'visibility factors' 't'e - ' (dotted lines) for two values of xe, the
fraction of the critical density in the form of ionized plasma (assumed
independent of z for illustrative purposes). Even a small amount of reheating
is sufficient to produce a last scattering surface t hat is more 'smeared out' in
redshift than the standard model of recombination (labelled 'rec'). In some
models for galaxy formation where energy (and maybe heavy elements) are
generated at large redshifts, other kinds of opacity may be competitive with
Thomson scattering (from H ogan, Kaiser and Rees, 1982).
2 / Xe = 1 Xe = 0.00 1 rec
r
1 I I
/ /
/ . ... . . . ..
/ ./·
/. . . . . . . . . ..
0
� .. ..
. .
. .
.
· ·. . .
. . .
. .
1 10
.
1 00 1 000
l +z
were in gas which had been reionized at z > 10), the horizon scale subtends
� 1 5°. Consequently the high isotropy of the background (�T/T< 10 - 4 )
implies that, if no � 1
£ ,.....
•
< l 0 _ 4 ior {
l > L/ctH > 1 (z * � lO)
1 > L/ctH > 310 (z * � 10 3 ) .
(3.3)
There seem to be no ambiguities in this argument; there are only minor

changes if other values of n0 are taken. Fluctuations on scales ctH would
-
cause large angle (quadrupole and octopole) effects ; the smaller angular
scale observations probe smaller values of L (cf. (3 . 1)).
[If n0 ::::;; 1 then our particle horizon will eventually grow to encompass an
infinite amount of matter that we cannot yet observe. Inhomogeneities on
scales � ctH can induce gradients and shear across the region within our
horizon , and may cause anisotropy in the microwave background. The
observed limits to dipole and quadrupole anisotropy can thereby be used to
set some limits on these still larger scales (Grishchuk and Zel'dovich , 1978 ;
Kaiser, 1982). These limits, which are themselves somewhat model
dependent, become progressively less stringent (in terms of c) on larger
scales. We certainly cannot rule out a gross deviation from homogeneity
(bubble wall? M inkowski space?) on scales � 10 2 ctH . There can only be
'philosophical' reasons for extrapolating the observed inhomogeneity of the
part of the universe we can now study to a domain which (if the universe is
open) is infinitely larger.]
The scales to which (3 .3) applies are larger than those on which we
observe the most conspicuous inhomogeneity in the distribution of galaxies.
It would be of particular interest in constraining theories of galaxy
formation if we could observe temperature fluctuations due to the same
scales which now display clustering. The upper limits to � T/T on the
relevant scales of a few arc minutes (cf. (3 . 1)) are at the level of a few
times 10 - s.
On angles corresponding to scales L > L * we are basically just seeing
'metric fluctuations' which have been unaffected by pressure gradients and
have evolved acausally throughout the (post-inflationary?) Friedmannian
expansion phase of the universe. But if L < L* various complications arise.
(i) The Sachs-Wolfe contribution to �T/T may not be the dominant one.
Doppler effects due to peculiar motions may be even more important;
so also may effects due to the changes in the recombination and
decoupling time in perturbed regions.
(ii) Pressure gradients and damping may have affected the perturbations
478 M. J . Rees
on scales < l* at epochs before decoupling . This complicates the

relationship between t: and ((Jp/p)z = z* for these scales.
(iii) On the scales below (!lz* /z* )l * the temperature fluctuations are
smeared out by the gradual nature of the decoupling - i .e. the fact that
the last scattering surface is not sharp compared with the scale being
studied . Fu rthermore, these scales would be attenuated by photon
viscosity during recombination; this further complicates the
relationship between 1:1 T/T and the value of �p/p which survives after
decoupling . (In the case of no reheating , these effects due to smearing
and viscosity are important only for l < ctH/300.)
There is an extensive literature attempting to quantify the effects (i)-(iii)
above : to relate t:, !l T/T and the post-decoupling ((Jp/p) on the scales
relevant to cluster formation (Press and Vishniac, 1980; Peebles, 198 1 ,
1982 ; Silk, 1982 ; Bond and Efstathiou , 1986, 1987) . Most calculations are
predicated on the assumption that there is no early reheating , and that the
'cosmic photosphere' is determined by the recombination behaviour of the
primeval H-He plasma. This work offers a consistency check on scenarios
which preclude sufficiently early reheating ; but it is important to stress that
the interpretation of 1:1 T/T isotropy limits on small angular scales is very
sensitive to the value of z * and !lz * . Under more general assumptions,
smearing effects can be very large for scales l ;:::; 10 2ctH ; furthermore, any
-
anisotropies actually observed on small angular scales may be of 'secondary'

origin, arising from reheated gas, dust, or discrete sources rather than being
directly related to t:. (See Hogan , Kaiser and Rees ( 1982) and Bond, Carr
and Hogan ( 1986) for further discussion of secondary fluctuations.)
Even 15 years ago, microwave background isotropy measurements
seemed tantalisingly close to the sensitivity level where they were bound to
reveal some effects. This still seems so . If they could be pushed below the
10- 5 level , constant curvature fluctuations with any value of t: could be
excluded (see Fig. 10 .5) , unless one adopted a quite different pictu re for
galaxy formation. The microwave fluctuations on very small angular scales,
incidentally, which are harder to interpret because they are sensitive to the
thermal history of the intergalactic medium after recombination and to
discrete (perhaps pregalactic) sources, can for this reason serve as probes for
the 'dark age' between 106 and 109 years, during which the first structures
developed .
10.4 Formation of protogalaxies
There are, as described briefly in Section 10.2, at least three senous
candidates for the dark matter in galactic halos and clusters : low mass stars ;
black hole remnants of very massive objects ; or non-baryonic matter,
perhaps in the form of supersymmetric particles or axions. I would myself
lay similar odds on these three options at the moment. However, it is
gratifying that we can expect the odds to change quite rapidly, owing either
to improved observational and experimental searches, or to progress in
particle physics .
Fig. 10.5. This diagram, from Rees ( 1 983), depicts the astrophysical limits
on the amplitude t: of adiabatic metric fluctuations, on various scales I. On
large scales, the microwave background isotropy offers stringent upper
limits. On very small scales, t: must merely be not so large that too much of
the universe collapses as the relevant scales enter the horizon ; the absence of
distortion in the microwave background sets a slightly better limit for mass
scales 1 04- 10 1 3 M 0 . The requirement that bound systems have 'turned
around' by the present epoch gives lower limits (also plotted). A spectrum
with t: � 10 - 4 on all scales is acceptable if the universe is dominated by non
baryonic matter which can start clustering before tree · If t: has a power-law
dependence on I and is - 10 - 4 on the scales relevant to galaxy formation,
0
then it cannot fall off more steeply than oc / - · 1 5 without causing excessiv�
production of primordial mini-holes . Also plotted (dotted lines) are the
amplitudes t:8 of primordial gravitational waves that can, within the next few
years, be probed by : (a) doppler tracking of spacecraft , (b) timing of 'quiet'
pulsars, and (c) timing of the orbit of the binary pulsar. This diagram is
drawn assuming a 'llat' background universe with n 0 1 . If n 0 < 1 the limits
on large scales are modified . I ndeed the appropriate definition of t: is
=
ambiguous on scales exceeding the Robertson-Walker curvature radius.
1 0-2 1 02 1 06 1 01 0 - (light years)
�
............ ........... . .. . ) ....
�.
·
(a)· ...·...
0
•
.• ••
� \\ \� �·-'"
-.:\. '
-;- ' ' ''
- . 1 \ \ ....... . ·�
•
·� (c
€
'\ \.
� ' ' ' ' ' �:::.,....

§
un ord 1a
Limits on pr Spec tral distortio
ole s \. ns
�
lack h
(b)
b of background
�
1 0-2 .l;Jfi
Galaxy
{ �
w ithout ' inos'
' inos ' '----.--"

1 0 -4 formation w ith ::::.. ::::.. ::::.. ::::.. :::..
'\\ \\\ "
If no early reheating
1 0-e 1 0-4
480 M . J . Rees
A separate line of attack that might shorten the list of candidates entails
exploring the implications of each for galaxy formation - specifically, for
processes whereby small primordial perturbations evolve · into proto
galaxies and clusters. The key parameter here is the spectrum of the density
fluctuations at the era of recombination, which is determined by initial
conditions (described, for instance, by (3 .2), modified by damping processes
during the fireball phase (t � 10 6 years)). The mass-scale of the first bound
system (i.e. the scale on which the growing perturbations after
recombination first become non-linear) roughly corresponds to the scale at
which the fluctuation spectrum peaks. Fig. 10.6 depicts three rival
cosmogonic schemes (see caption for further details). The disparity between
them is highlighted if we envisage what the universe would have been like,
according to each, at the epoch corresponding to z < 10. In hierarchical
'bottom-up' schemes, a lot could have happened already. In others , the
universe is then still amorphous neutral hydrogen , with all parts expanding
in nearly undeviated Hubble flow. The cosmological 'dark age' starts when
microwave background photons shift longward of the visible band , when
the universe is 10 6 years old : it ends when the first bound systems (or their
constituent stars) light up. We know that quasars - and so, presumably ,
Some galaxies ...:. had formed at Z "-' 4 (t - 109h - l years if 0 0 = 1), but this is
merely a lower limit to the redshift of the first non-linearity. We do not know
whether the 'dark age' is brief, or lasts for a billion years. We are more
confused and ignorant about this phase of cosmic history than many seem to
be about the first 10 - 3 5 s.
A related cosmogonic issue is this. To what extent is the present large
scale structure a rather direct consequence of initial fluctuations, imprinted
at 10 - 3 5 s? Or does it, contrariwise, result from secondary perturbations
generated or seeded by the first bound systems? Hogan ( 1986) has suggested
tlie phrase 'paleogeny versus neogeny', to denote this dichotomy.
1 0 .4 . 1 The 'cold dark matter' cosmogony

The three schemes depicted in Fig. 10.6 have all been reviewed elsewhere
(see, for instance, Efstathiou artd Silk , 1983). Here I shall focus on the
specific cosmogony which postulates that the hidden mass is so-called 'cold
dark matter' . In other words, this material is envisaged as weakly interacting
non-baryonic stuff (e.g. GeV supersymmetric particles or axions) whose
thennal motions are too low for free streaming to damp out fluctuations on
any significant scale. (In this respect, it should be contrasted with a model
where light ( "'"' 10 eV) neutrinos dominate : these would be moving
Galaxy formation and dark matter 48 1
Fig . 10.6. The cosmogonic processes after (re)combination depend on the

spectrum of fluctuations which have survived damping processes, etc . , at
earlier times . After tree , linear g rowth proceeds roughly according to the law
(�p/p) oc t213, and the first gravitationally bound systems to form will have the
mass for which the density contrasts at tree are biggest . In case ( 1 ) ,
(super)clusters are the first systems to condense, a n d they form quite recently .
Indeed , models of this kind run into difficulties because we observe quasars
with redshifts z as large as 4, whereas if superclusters had collapsed as early
as this, they would now be denser, and display a higher density contrast,
(caption continued overleaf)
t (yr) T (K)
' Pancake ' + Hierarchy Hierarchical
fragm entation with ' Crosstalk ' st ructure
1 0 10 3
I
.. ..
I
I
} Quasars
I
I
I
1 09 I
I
I
- - - - - -·- - - - - - - - f- -+ - -
.. I
30 - - - - - - +--
H I f
I
T
I I
• I
I I
I I
I I
I
I I
I I
:
I
I :
I
300 - - - - - - .- - - - -•, - ,I - , -- - - - - � - -+ - -
I
.
I
I
I
1 07
I
.
M
I
I
T
I I I
I I
I
I
I I
I
I
I
I I
1 06
••
I I
I
I ..
: : : 1
3000 11l1T1 TT i1 1T1l1 TT ill TT Ill TT iTITT IT1Ti !11 11 1 /i 111 TT 11j il 11.i iT i)1 TT
I I
: • I
:. (Re)combination : :. :
.. .. ..
.
0. 1 0. 1 0. 1
0. 0 1 0.0 1 0. 0 1
0.00 1 /".... 0 . 00 1 0.00 1

---�--- M M M
i o 1s
'--'--�-�- ---�---
1 05 1 0 1 0 1 0 1 5 M 0 1 05 1 0 10 M0 1 05 1 0 10 1 0 15 M0
CD Adiabatic: � Adiabatic: Q) Power law entropy

Baryon-dominated Dominated by axions perturbations in
or v-dominated or heavy (' cold ') baryon distribution
' . . . inos '
Small scales attenuated
by photon viscosity or No damping, but growth
free-streaming of v of small scales
inhibited during radiation
dominated phase
Flu ctuations scale-independent on entering

horizon (Harrison/Zeldovich spectrum)
482 M. J . Rees
Fig . 10.6 (continued)

than is seen . In case (2), baryonic systems of - 106 M 0 condense in potential
wells produced by 'cold dark matter' or 'inos', which are presumed to be slow
moving, so that they are not homogenised on small scales as neutrinos are.
(This specific case is discussed more fully in the text.) The third case shows the
spectrum that might arise from primordial entropy perturbations. In cases
(2) and (3), subgalactic systems would form before galaxies; if these
subgalactic systems provided an energy input, they could in principle
generate 'secondary' perturbations on larger scales which could swamp the
genuinely primordial ones.
relativistically until the stage in the cosmic expansion when k T fell below
......, lO eV, with the consequence that scales � 10 1 5 M0 are homogenised by
free-streaming, leading to a cosmogony where superclusters form first and
then have to fragment into individual galaxies.)
It is relatively straightforward to calculate the growth factor for density
perturbations in 'cold dark matter' between the era when they enter the
particle horizon and the present. All scales grow roughly as the scale factor
(i.e. (bp/p) oc R oc r - 1 oc t213) after the epoch when the universe becomes
dynamically dominat �d by dark matter. This result can be derived by simple
arguments bas·ed on Newtonian cosmology (McCrea and Milne, 1934).
However, at earlier times the universe is dominated by photons, and the
expansion timescale is proportional to (GP rad ) - 112; perturbations in the
dark matter alone with a growth time (GP co M ) - 112 do not then have time to
grow significantly. See Fig. 10.7. This effect, termed 'stagspansion' by
Blumenthal and Primack ( 1983), was first calculated by Guyot and
Zel'dovich ( 1970) and Meszaros ( 1974). If the initial fluctuations have the
Harrison ( 1970)-Zel'dovich ( 1972) scale-independent form , so that each
scale has the same amplitude e when it is first encompassed within · the
particle horizon (i.e. at a time oc M), the present-day fluctuations would be
proportional to M - 213 for large masses, and be essentially independent of M
at low mass. The spectrum flattens at low masses because all scales entering
the horizon before the epoch of equal densities undergo essentially the same
growth. The resultant spectrum of density perturbations actually has a
rather gradual 'rollover' in the transition region (Fig. 10.8). There is
nevertheless a characteristic mass imprinted by cosmology, which is smaller
than the mass now within a 'Hubble volume' by a factor (Qrad/Q0 M )3'2,
where n rad ( ......, 10 - 4) is the present contribution of primordial radiation to n.
The amplitude of the fluctuations (i .e. the value of e in (3 .2)) cannot yet be
reliably predicted theoretically. The vertical scaling in Fig. 10.8 can
therefore be normalized to ensure that (( fJ p/p ) 2 ) 11 2 be now - 1 on a scale

"' 5h - 1 Mpc. (This is inferred from the clustering properties of galaxies; the
normalization would need some adjustment if, as discussed further m
Section 10.5, the galaxies do not trace the overall mass distribution.)
Fig. 10.7 . The growth of adiabatic fluctuations in a universe dominated by

'cold dark matter'. The mass of cold dark matter within the particle horizon
is shown as a function of time on a log-log plot. For t > teq (corresponding to
the redshift indicated on the figure), all scales grow at the same rate. Before
teq , when the expansion is radiation-dominated, there is essentially no
growth, because the growth timescale much exceeds the expansion timescale.
If the fluctuations on all scales enter the horizon with equal amplitude (the
'Harrison/Zel'dovich' hypothesis) , then the present-day spectrum would
have the approximate form written on the right-hand side. (The accurately
calculated spectrum is shown in Fig. 10. 8 .) The cold dark matter
perturbations start to grow at teq , whereas radiation pressure inhibits growth
of baryonic fluctuations on relevant scales until the (later) recombination
time tree · Cold dark matter fluctuations thus have a 'head start' (baryons
being able to fall into the resultant potential wells after t,ec ); this permits an
acceptable cosmogonic scheme with lower fluctuation amplitude t:, and
smaller microwave background fluctuations, than in a baryon-dominated
universe (cf. Fig. 10.5).
M H(CDM)
§£ a:
p - Af - t
\
0
Supercluster
Cluster I
=
- ---i
: f> p a: '7'- ! a: tt
I p
Galaxy ' -r
-
� [ [ No damping of ] -
= I � � Const
__
small scales
I
No significant
growth
feq tree tnow

t
( ( 1 + z) � 1 04 S"2h �o 1
484 M. J. Rees
Because the spectrum in Fig. 10.8 is so nearly flat for small M, the typical
fluctuation of 106 M 0 would collapse no earlier than the epoch
,....,
corresponding to z � 10. The build-up of structure is hierarchical , in the

sense that smaller scales tend to form earlier. However, because of the flat
spectrum, there would be complicated 'crosstalk' between many different
scales (cf. middle panel of Fig . 10.6) : the 30' peaks in the density distribution
on 10 1 1 M0 scales would have the same amplitude as typical peaks of
mass 106 M0, and would therefore collapse at the same time. It is
,....,
consequently hard to analyse (either analytically or numerically) even the

purely dynamical and non-dissipative aspects of the clustering. However,
those studies that have been done (e.g. Davis et al. , 198 5 ; Frenk et al. , 1985)
are encouraging , in that , when the amplitude of the fluctuations is
normalised so as to match the data on galaxy clustering, the finer-scale
disposition of the dark matter closely reproduces the sizes and profiles of
galactic halos (see Fig. 10.9).
1 0 .4 .2 The first stars

Pressure gradients in the baryons would prevent bound systems of ;:5 10 5 M0
from collapsing, even after recombination . I f there were n o primordial
perturbations in the baryon-photon ratio (i .e. the initial fluctuations were
Fig. 10.8. The root-mean-square fluctuations within a randomly-placed

sphere containing mass M for cold dark matter, for two specific models
calculated by Blumenthal and P rimack ( 1 983) (adapted from Blumenthal et
al. ( 1 984).
6 8 10 12 14 16 18
log M/M0
adiabatic), the first bound baryonic systems would be associated with high-u
peaks in the dark matter distribution on mass scales 105-10 6 M 0 . These
would collapse at a redshift z of order lOv , where v is the number of standard
deviations above the mean. Everything is straightforward while all
perturbations remain linear, but once the first bound baryonic systems have
formed - once 'first light' occurs - what happens next could be crucially
influenced by feedback from these systems. This influence depends on the
mass spectrum of the first-generation stellar objects.
Our poor understanding of early star formation impedes quantitative
study of all galaxy formation schemes, particularly the hierarchical ones
where the first stars form at pregalactic epochs. We cannot reliably decide
Fig. 1 0.9. This shows three stages in the evolution, at redshifts z = 2.5, 1 .0 and
0 respectively, of an N-body computer simulation intended to represent the
contents of a cubical volume of present dimensions 14 h - 1 Mpc in a universe
with n = 1 (Frenk et al., 1985). The initial fluctuations were chosen to match
those expected for cold dark matter (see Fig . 10.8). When the amplitude is
chosen so as to match the present-day scale of clustering there is no further
freedom in the model ; it is therefore gratifying that the matter is aggregated in
systems whose properties match those of galactic halos . The baryonic
component, which settles dissipatively in these halos to form the luminous
content of galaxies, is unimportant for the overall dynamics . The present
structure is th� outcome of gravitational instability, in an initially almost
homogeneous expanding universe, developing into the non-linear regime. In
a celebrated letter to Bentley, Newton wrote 'It seems to me, that if the matter
of our sun and planets and all the matter of the universe were evenly scattered
throughout all heavens, and every particle had an innate gravity toward all
the rest, and . . . if the matter were evenly disposed throughout an infinite
space, it could never convene into one mass; but some of it would convene
into one mass and some into another, so as to make an infinite number of
great masses, scattered at great distances from one another throughout all
that infinite space. And thus might the sun and fixed stars be formed . . . '
. . j .
.' '{- .
-�·
·' -
·,
< : ,_ ... ;
�-
z = 2.5 z = 1 .0 z = O
486 M. J . Rees
whether a collapsing baryonic cloud of ,...,, 105 M 0 will turn into a single very
massive star, or fragment into 107 stars of ultralow mass. In the latter case,
the first stars would not inject any significant energy into the remaining gas.
On the other hand, if the first stars were ordinary massive stars or
supermassive objects, their ultraviolet radiation would photoionize the
medium. Indeed, only 10 - 4 of the initial mass need turn into such stars in
,...,,
order to generate enough UV radiation to photoionize everything else

(Hartquist and Cameron, 1977). An additional possibility is that heavy
elements generated by pregalactic supernovae may be spread through the
intergalactic medium at an early stage. Details of these feedback processes
are discussed by Couchman and Rees ( 1986). The baryonic Jeans mass
would be raised from 105 M0 to 108 M 0 , with a consequent pause in
,...,, ,...,,
the cosmogonic process until systems exceeding this larger mass undergo
collapse; moreover, no more than 10 - 4 of the baryons (i.e. the � 3a peaks)
need condense into massive stars before reheating chokes off any further
collapse of 105 M 0 systems.
,...,,
Gas would subsequently condense into dark matter potential wells in the
mass range 108- 101 2 M 0 . The lower limit is set by the Jeans mass after
reionization, and the upper is the maximum scale where bremsstrahlung and
hydrogen and helium recombination cooling is efficient enough (see Fig.
10. 10; Compton cooling would allow larger and hotter clouds to collapse at
z � 10) . The luminosity function of galaxies depends primarily on two
things.
(i) We need to know the mass distribution of isolated virialized systems of
dark matter. This is a hard problem when the initial fluctuations have
the spectrum shown in Fig. 10.8 because of the cross-talk between
various scales, and its solution awaits N-body simulations with larger
dynamic range than have so far been carried out. At the moment it is
unclear whether well-defined substructure survives, or whether bound
systems on scales of, say , 108 M 0 are rapidly engulfed in a larger system
before baryons have much chance to condense within them.
(ii) The luminosity of a galaxy resulting from infall into a given potential
well depends on what fraction of infalling gas is retained in each
virialized clump of dark matter (likely to be larger for deeper potential
wells, i.e. those of large mass and/or those which evolve from high-a
peaks) and on the kind of stars it turns into.
The above two issues will need to be settled, before we can reliably
estimate the fuminosity function of galaxies, even when our starting point is
the specific assumption of a cold dark matter spectrum that evolved from
Harrison/Zel' dovich initial fluctuations.
The following are among the observations that would help to decide
among the three cosmogonic schemes illustrated in Fig. 10.6, and to test the
'cold dark matter' model in particular.
(i) The upper limits on the microwave background fluctuations on small
angular scale constrain the amplitude of ( (6p/p) 2 ) 1 12 at the
Fig . 10. 10. This diagram , adapted from Rees and Ostriker ( 1977), delineates
the mass-radius relation for a self-gravitating gas cloud whose cooling
and dynamical timescales are equal (assuming cooling d ue only to
bremsstrahlung, H and He recombination , and line emission). A cloud of
given mass whose radius was initially very large would deflate quasistatically
(because tcool > tdyn ) until it crossed the critical line; it would then collapse in
free fall and could fragment into stars. This simple argument (which can
readily be modified to allow for non-spherical geometry, a non-baryonic
component of mass, etc.) suggests why, irrespective of the cosmological
details, no galaxies form with baryonic masses > 101 2 M 0 and radii
> 10 5 pc . We would l ike to understand why galaxies have these observed
masses and radii to the same extent that we understand the dimensions of
individual stars. The order-of-magnitude considerations summarised in this
diagram are probably part of the story, but to fill in the details we need to
know more about the initial fluctuations, and also about the efficiency of
star formation in protogalaxies.
I O'
1 06
1 04
1 010 1 0 12 I 0 14
M/M"'
488 M. J . Rees
recombination epoch, and are already sensitive enough to exclude some

models. Uson and Wilkinson ( 1984), for instance, find ( (� T/ T) 2 ) 1 ' 2 <
2 . 8 x 10 - 5 for angles "' 3.5'. A further improvement, even by a factor no
more than 2, could be crucial. Measurements on larger angular scales
are directly probing the metric fluctuation, and could test the
consistency of a Harrison/Zel' dovich spectrum where this amplitude is
scale-independent.
(ii) It would be valuable to improve data on clustering correlations (and
non-Hubble velocities) on scales > 1 0 Mpc. Because the cold dark
matter fluctuation spectrum falls off steeper than a 'white noise'
spectrum at long wavelengths, the correlation function in this model
should go negative for scales > 1 8h - 2 Mpc, unless (for some
astrophysical reason) the galaxy distribution does not trace mass on
these scales.
(iii) One would like to know whether there is dark matter in dwarf galaxies.
The observational evidence .here is uncertain , because the stellar
velocities are below 10 km s - 1 (Fig. 10.2). However, such dark matter
certainly could not be light neutrinos, because they would need a higher
phase space density than they had in the big bang (Tremaine and Gunn,
1979; Madsen and Epstein, 1984); cold non-baryonic matter would be
an acceptable possibility, however.
(iv) Most crucial of all would of course be firm evidence for particles that
could constitute 'cold dark matter'. Experimental hopes for axion
searches, and for the detection of supersymmetric particles, are by no
means fanciful (see Section 10.2).
As a final comment, note that most discussions of galaxy formation
postulate initial Gaussian fluctuations (of the kind that might survive from
the quantum or inflationary era). Moreover, attention has focussed on so
called 'adiabatic' perturbations, in which the photon-baryon ratio is
unperturbed, because GUT models suggest that this ratio may be fixed by
physical processes controlled by the expansion rate at very early times. But
there are other options : for instance, it could be that cosmic strings provide
the seeds around which galaxies and clusters condense (Vilenkin, Chapter
1 1 of this volume).
10.5 A flat (Q0 1 ) universe?
=
In an influential review paper published more than a decade ago, Gott et al.
( 1974) summarized the evidence bearing on n. They concluded that the
dynamical evidence favoured a value 0. 1---0 . 2, and noted that if the matter
were all baryonic, the lower end of this range was compatible with the value
favoured by standard big-bang nucleosynthesis (for a Hubble time tH � 2 x
101 0 yr, a value consistent with the ages of globular clusters, etc.). Much new
evidence has accumulated since 1974, especially on cluster dynamics and
element abundances ; and some relevant theoretical issues have been refined
and elaborated. But, if one were to update Gott et al. ' s discussion , their net
conclusion would not change much .
There has, however, been a marked change in theorists' attitudes. This is
partly because non-baryonic matter is now taken much more seriously, and
seems in some ways almost a natural expectation . But the main element in
the discussion is the concept of 'inflation' : this is so appealing , and resolves
some well-known and stubborn cosmological paradoxes in such a natural
way that it instils a strong prejudice in favour of n o = 1 . It is perhaps worth
'
spelling out the basis for this prejudice.
For all the present observable universe to have evolved from a region that
was in causal contact at the earliest times, inflation by a factor of at least 10 3 0
is required. In most versions of inflation the exponential growth, once
started, rapidly continues for many expansion timescales : it is likely to
overshoot, stretching any small part of an initial chaotic hypersurface so
that it becomes essentially flat over our present horizon scale. This would
yield n o = 1 , with a precision of order 1 part in 10 4 (the expected fluctuation
3
amplitude) . For inflation to yield the dynamically preferred value n = 0. 1 or
0.2 the inflation factor would have to be 'just' 10 0 , making the present
Robertson-Walker curvature radius of the order of the Hubble radius. This
would demand some coincidence. But there would then be an additional
requirement that appears still more contrived : our presently observable part
of the universe would have to arise from a segment of the initial hypersurface
with the seemingly very special property that its curvature was uniform to a
few parts in 105 ; otherwise the curvature fluctuations that could induce
quadrupole effects in the microwave background would not be at least 104
times smaller than the overall Robertson-Walker curvature (Wilkinson,
1986). Our universe could thus not have inflated from a typical element of an
initial chaotic hypersurface : if n # 1 , the required region would have to be
special, rather as a sphere would seem specially smooth if its surface
irregularities amounted to 10 - 5 of the uniform mean curvature .
10.5 . 1
Inferring Q0 from the cosmic deceleration
In Friedmann models with zero cosmological constant, n is directly
proportional to the deceleration of the cosmological expansion.
490 M. J . Rees
Programmes attempting to measure this deceleration have been pursued

throughout the last 20 years, but there is now not much optimism that they
can succeed until we understand better how galaxies evolve. ·
There is a spread in the properties of galaxies, just as there is among
individual stars. One needs, therefore, to find some class of galaxies which
serve as 'standard candles', so that their distance can be inferred from their
apparent brightness. Otherwise there will be a big scatter in the magnitude
redshift plot. The brightest galaxies in clusters are the best candidates, but
even they display a 30 per cent scatter. Another problem is that galaxies
become almost undetectably faint, even when observed with the largest
telescopes, when they are far enough away that the effects of the deceleration
should really show up.
But the worst problems stem from galactic evolution. The galaxies seen at
large distances are systematically younger than nearby ones. Even if a
certain class of galaxy provided precise standard candles at the present , one
needs to know how each candle changes as it burns. The galaxies that are
crucial for cosmological tests are those whose light has been journeying for
at least 5 billion years towards us, which are therefore being seen at only
about half their present age.
There are two aspects to the evolutionary correction. In a younger
elliptical galaxy many stars would be shining which by now have died, and
the present stars would be seen at an earlier stage in their evolution . This
alters the brightness and the colour of the galaxy, the trend being that a
younger galaxy would appear somewhat brighter than its present-day
counterpart. But there is a secondary evolutionary correction stemming
from the fact that a galaxy is not a self-contained isolated system. We can see
many instances where galaxies seem to be colliding and merging with each
other; and in rich clusters the large central galaxies may be cannibalising
their smaller neighbours. (In a few billion years this may , incidentally,
happen in our own local group. The Andromeda galaxy is falling towards
our own M ilky Way, and there may be a collision between these two large
disc galaxies, the likely remnant being a bloated amorphous 'star pile'
resembling an elliptical galaxy.) Many big galaxies, particularly those in the
centres of clusters, may result from such mergers, traces of disturbance
having by now been erased. This process would obviously result in big
galaxies, having been , on average, fainter in the past. There are thus two
evolutionary effects, both uncertain (and with opposite signs) , either of
which could be large enough to mask the geometrical difference between a
universe with n o = 1 and one with n o � 0.2.
A new cosmological test, involving the number of galaxies of a given type

as a function of redshift, has recently been carried out (Loh and Spillar,
1986). This is still somewhat vulnerable to evolutionary effects but obviously
utilises more information than simple number counts of galaxies. It yields
results consistent with n0 as high as unity.
Note that the cosmic deceleration R is related to n via R R/R 2 n;2
- =
only if the Friedmann equations apply, if the cosmical constant is zero, and if
the dominant form of mass-energy has an associated pressure � -�pc 2 •
Ideally, therefore, one would like to determine n and R independently, as a
test of these assumptions.
1 0 . 5 . 2 'Biased' galaxy formation
The dynamical evidence from clusters and from galactic halos (surveyed in
Section 10.2) does not offer evidence for any value of n0 higher than "' 0.2. If
n0 is indeed unity, then 80 per cent of the mass is unaccounted for even by
dynamical considerations : it is not just the light, but evidence for gravitating
matter itself, which is 'missing'. This raises the interesting question of
whether the dynamical evidence is nonetheless compatible with n0 = 1 .
If n0 = 1 , then the dominant mass must not participate fully in the
observed clustering : galaxies must be more 'clumped' than mass in general,
so that their spatial distribution enhances and exaggerates the
inhomogeneity on large scales. This requires some kind of 'bias' in the
formation of galaxies. There are three ways this would come about (see
Dekel and Rees, 1987, for fuller discussion):
(i) The entire universe may be pervaded by a uniform component of
'missing mass' of a different nature from the clustered dark matter,
contributing n "'0.8.
(ii) There may be one important kind of dark matter, but the baryonic
component may be segregated from it even on scales - 30h - 1 Mpc, so
that galaxy formation occurs only in certain regions. This could result
from large-scale fluctuations in the initial baryon/photon ratio, from
gas dissipation within superclusters of collisionless dark matter (e.g.
neutrinos), or from very energetic winds or blast waves pushing the gas
over large distances.
(iii) Less extravagant in energy is the possibility that the large-scale baryon
distribution does trace dark matter on scales :<: lh - 1 Mpc, but the
efficiency with which baryons turn into luminous galaxies is modulated
by large-scale environmental effects.
The universe may be dynamically dominated by 'ultrahot' weakly
492 M. J . Rees
interacting particles which do not cluster because they have velocities

103 km s- 1 (alternative (i) above). But if the 'ultrahot' particles are relics of
an early epoch, their mass would have always been dynamically dominant
over the baryons, and would have inhibited gravitational clustering
altogether. This would also yield an unacceptably fast expansion timescale
during nucleosynthesis. A way round this difficulty - though maybe a rather
contrived one - involves supposing that these particles arise from non
radiative decay of heavy particles with lifetimes only slightly shorter than the
age of the universe.
The universe could be flat with n0 < 1 if a non-zero cosmological constant
contributed to the curvature such that n0 + 1At� = 1. In some respects this
idea resembles 'ultrahot' particles, but it has the advantage that the A -term
is unimportant at early epochs so it would not have had such a serious
inhibiting effect on galaxy formation. However, for the A-contribution
today to be comparable with the ordinary matter, the required fine-tuning is
as ad hoc as the one we intended to avoid by adopting n0 = 1 .
A wide class of possibilities in category (iii) involve the idea that galaxies
formed only from exceptionally high peaks of the initial (fluctuating) density
distribution. If the local distribution function of bp has a steeply decreasing
tail, like a Gaussian, high peaks occur with enhanced probability in the
crests rather than the troughs of a large-scale fluctuation mode, and hence
display enhanced clustering (if the distribution function has a flat tail the
effect is weaker and it might even be reversed). See Bardeen et al. ( 1986) for a
full discussion of the theory.
The high-amplitude peaks would collapse earlier, and have higher density
at turnaround, than more typical fluctuations on a given mass scale. This
could, in principle, by itself account for the biasing. For instance, collapse
and fragmentation of gas within protogalaxies is highly sensitive to the ratio
of the cooling time ( oc p - 1 for bremsstrahlung and recombination cooling)
to the collapse time ( oc p - 1 12 ) : it occurs more rapidly when this ratio is small.
Moreover, if the high-amplitude perturbations collapsed at z � 10,
Compton scattering of the microwave background would guarantee
efficient cooling ; this process, whose efficiency is proportional to ( 1 + z) 4 ,
would be unimportant for lower-amplitude systems collapsing at late times.
Protogalaxies that formed from high-er fluctuations would have higher
velocity dispersions (and escape velocities) than later-forming systems of
similar mass. This makes it easier for them to retain gas that might otherwise
be expelled by, for instance, supernova-driven winds (Dekel and Silk, 1986).
The formation of galaxies from lower amplitude perturbations may be
fu rther impeded owing to some kind of negative feedback induced by the

first galaxies, suppressing the baryonic infall or affecting the stellar initial
mass function (Rees, 1985 ; Silk, 1985).
1 0.5.3 Biasing in the cold dark matter (CDM) cosmogony

The standard CDM cosmogonic scheme (!l 0 = 1 and scale-invariant initial
curvature fluctuations) cannot be reconciled with observations without
some form of biasing . Detailed simulations of such models, hypothesizing
that galaxies form only from regions which are > 2.5a peaks of the initial
mass distribution (smoothed on a galactic scale) , show a good fit with the
galaxy-galaxy correlation function, the M/L ratio in clusters and the Virgo
infall (Davis et al., 1985). Although this is encouraging, we still need a
genuine physical justification for the biasing.
The CDM cosmogony will not be able to predict the luminosity function
of galaxies until we develop a physical understanding of several distinct
phases of galaxy formation :
(i) Baryons are presumed to condense in virialized 'halos' of dark matter in
the approximate mass range 10 8- 10 1 2 M 0 . For larger masses,
dissipative cooling may be inefficient ; below ,..., 10 8 M0 the potential
wells may be too shallow to capture primordial gas (especially if this gas
has been ionized and heated to ,..., 10 4 K). The mass distribution of
isolated virialized systems can in principle be learned from N-body
simulations. In practice, progress is impeded by the large dynamic
range involved ; galactic-mass halos build up hierarchically, via a
complicated series of mergers, between z � 5 and the present epoch
(Frenk et al. , 1985).
(ii) Even if the dissipationless clustering of the dark matter were accurately
known , the fate of the baryonic component - how much gas falls into
each potential well, and how much is retained - involves complex gas
dynamics.
(iii) The resultant luminosity and brightness profile of the galaxy depends
on the IMF of the stars that form from the gas, and on the efficiency of
star formation .
The biasing could in principle involve any (or all) of these three aspects of
galaxy formation. The N-body simulations suggest that even stage (i) may
lead to substantial biasing : the dark halos associated with bright galaxies -
those with circular velocities vc > 200 km s - 1 - are themselves more
clustered than the overall mass distribution, essentially because the binding
energy of galactic mass perturbations is significantly affected by whether
494 M. J. Rees
they are in a peak or a trough of the fluctuation spectrum at larger

wavelengths. (The growth of the perturbation on galactic scales is boosted
or suppressed in surroundings whose overall expansion mimics an n > 1 or
an n < 1 universe respectively.) This effect alone might account for the
requisite biasing of bright galaxies, although it would not explain the
tendency of galaxies of low luminosity to cluster like brighter galaxies.
Nevertheless, any of the other mechanisms involving autonomous biasing
or feedback could also be operative, modifying stages (ii) and/or (iii) of the
formation process.
Very-large-scal.e structures. There are indications that the very-large-scale
superclusters and voids may call for a formation (and bias) mechanism over
and above the type of biasing process that could plausibly account for the
galaxy correlations on scales lOh - 1 Mpc. One crucial observation in
-
addition to the presence of big 'voids' is the enhanced superclustering of rich

Abell clusters over the galaxies (and the matter) characterized by a
correlation function which is positive at least out to lOOh - 1 Mpc (Bahcall
and Soneira, 1983; Bahcall, 1986), and by associated velocities of
- 1000 km s - 1 • Another puzzling finding of potentially great importance is
the bulk motion of a region - lOOh - l M pc in diameter around us relative to
the microwave background frame of reference with a velocity of
- 600 km s - 1 (Burstein et al. , 1987). This superstructure, on such large
scales, is hard to reproduce in any of the standard scenarios that assume
initial fluctuations with a scale-invariant spectrum and random-phases; it is
difficult also to reconcile with the upper limits on the microwave
background fluctuations on arc minute scales.
If the density fluctuations began as quantum fluctuations of a free scalar
field during the era of inflation they are indeed expected to be Gaussian, but
it is also possible that the fluctuations arose from a different mechanism, in
which they would not in general be Gaussian, and have non-random phases.
If the initial fluctuations had non-random phases, then the universe might be
inherently smoother in some places than in others : galactic-scale
fluctuations with amplitudes large enough to form bound systems could
then abound in some regions, but be entirely absent in large 'calm' patches,
the latter corresponding to 'voids'.
Biasing is not just an ad hoc idea introduced by theorists to save the
philosophically attractive 0 0 = 1 model when confronted with apparently
conflicting evidence. A bias is essential in order to understand even the gross
features of the large-scale structure, in particular the big 'voids' and the
superclustering of clusters, and to reconcile any of the cosmogonic scenarios
under current discussion with the observed universe. Therefore, the search
for an appropriate physical bias mechanism, and especially for confirming
observational evidence, are of great importance.
On the theoretical side, the bias mechanism is intimately related to the
cosmogonic scenario and the nature of the dark matter. Although some of
the proposed bias mechanisms may seem somewhat contrived, others are
very plausible physically : some or all of them must have affected galaxy
formation to some degree and they should be worked out in more detail. It is
evident that the notion that galaxies trace mass is an unjustified assumption.
Answers to the following observational questions would help to
distinguish between various options outlined in this section.
(i) How much diffuse gas is there in the voids?
(ii) Are voids empty of all types of galaxy, or only those types that are most
conspicuous? Any evidence that galaxies of different morphological
types display unequal degrees of clustering is relevant here.
(iii) Are there any galactic-mass dark halos with no luminous galaxy within
them? Such objects might be expected in the cold dark matter model
(assuming fluctuations with random phases) if biasing is indeed
important ; and could , if their core radius were small enough, account
for gravitational lensing of quasars even when no lens is visible.
(iv) Have the large-scale inhomogeneities in the galaxy distribution
(superclusters, etc.) given rise to substantial deviations from the Hubble
flow on corresponding scales?
I shall conclude by summarising two candidate cosmologies, both of
which now seem appealing in their different ways :
(i) The first model has nb = ntotal :::::: 0 . 1 5 (and a Hubble time tH = 2 x
10 1 0 yr). The baryons would be partly in 'luminous' form (stars and gas)
and partly in 'dark matter' (low mass stars and/or black holes).
(ii) Alternatively, theoretical predilection may dispose us to favour !l 101a 1 =
1 + 10- 4. (The small uncertainty would arise only from the initial
curvature fluctuations, which need to be 10- 4-10- 5 on the scale of
galaxies and clusters, and would, in the simplest models, extend with
similar amplitude up to the Hubble radius and even beyond .)
Conventional big bang nucleosynthesis then suggests nb :::::: 0 . 1 ,
nnon-baryon :::::: 0 .9, and introduces a new dimensionless ratio into
cosmology. The large-scale structure of the universe then involves some
kind of biasing in the formation of bright galaxies.
It may become easier to decide between the options when we have a better
knowledge of galactic evolution and star formation. It is still unclear
496 M. J . Rees
whether the luminous parts of galaxies result from infall into a preformed
halo composed of quite different stuff, or whether, contrariwise, galaxies and
their halos resulted from a single dissipative collapse process· whereby the
IMF gradually changed as contraction proceeded. Observations of quasars
back to redshifts z � 4 can now provide direct evidence of what the universe
was like at an era when galaxy formation was still going on. Numerical
simulations which can follow not only gravitational clustering (cf. Fig. 10.9)
but also the complexities of gas dynamics - shock waves, radiative cooling
and fragmentation - will soon become feasible.
The problems of large-scale cosmogony are so intermeshed that we will
not really solve any until the whole picture comes into better focus. For
instance, we cannot test theories of galaxy formation and evolution until we
understand star formation (and the possible role of active galactic nuclei) as
well as the initial fluctuations. The relevant physics is not specially recondite
- indeed most of the physics that astrophysicists need is in 'Landau and
Lifshitz' - and the most intractable phenomena are complex manifestations
of Newtonian gravity and dissipative gas dynamics. But when we ask where
the initial fluctuations come from or what the dark halos are made of, we
realise that even the most ordinary galaxies pose questions that may
transcend the physics we understand.
References
Alcock, C. and Fahri, H. ( 1985). Phys. Rev., D32, 1273.

Applegate, J. and Hogan , C. J . ( 1985) . Phys. Rev., D31 , 3037.
Applegate, J., Hogan , C. and Scherrer, R . ( 1 986). Preprint.
Bahcall , N. A. ( 1986). In Galactic Distances and Deviations from Universal Expansion, ed.
B . F. Madore and R . B . Tully . Reidel : Dordrecht.
Bahcall , N. A. and Soneira, R. M. ( 1983). Astrophys. J., 270, 20.
Bardeen, J., Bond, J. R . , Kaiser, N. and Szalay, A. ( 1 986). Astrophys. J., 304, 15.
Blandford , R. D . and Jaroszynski, M . ( 198 1). Astrophys. J., 246, 1 .
Blumenthal , G . , Faber, S . M . , Primack, J. R . and Rees, M . J. ( 1984) . Nature, 31 1 , 5 17 .
Blumenthal, G . and Primack, J . R . ( 1983). In Fourth Workshop on Grand Unification, ed .
H . A. Wilson et al., p . 256.
Bond, J . R., Carr, B . J . and Hogan, C. ( 1986). Astrophys. J., 306, 428.
Bond, J . R . and Efstathiou , G . ( 1986). Mon. Not. R. Astron. Soc., 2 18 , 103 .
Bond , J . R . and Efstathiou, G. ( 1987) . Rev. Mod. Phys. (in preparation) .
Bu rstein, D . , Davis, R . L . , Dressler, A . , Faber, S . M . , Lynden-Bell, D., Terlevich , R . J .
and Wegner, G. ( 1987). Astrophys. J. (in press).
Canizares, C. R. ( 1982). Astrophys. J., 263, 50.
Carr, B. J . , Bond, J. R. and Arnett, W. D. ( 1984). Astrophys. J., 277, 445.
Chang , K. and Refsdal , S . ( 1979) . Nature, 282, 56 1 .

Couchman, H . M . P . and Rees, M . J . ( 1986) . Mon. Not. R. Astron . Soc. , 221 , 53.
Cowsik, R . and McLelland , J. ( 1973). Astrophys. J., 180, 7.
Davis, M . , Efstathiou, G . , Frenk, C. S . and White, S . D . M . ( 1985). Astrophys. J., 292,
37 1 .
Dekel, A . , Einasto , J . and Rees, M . J . ( 1987) . Rev. Mod. Phys. (in press).
Dekel , A. and Rees, M . J. ( 1987) . Nature (submitted).
Dekel , A. and Silk, J. I . ( 1986) . Astrophys. J., 303, 39.
De Rejula , A. and Glashow, S . L . ( 1984). Nature, 312, 734.
Drukier, A. K . , Freese, K. and Spergel, D. N. ( 1986). Phys. Rev. , 033, 3495.
Dyer, C . ( 1976) . Mon. Not. R. Astron. Soc., 175, 429 .
Efstathiou , G . , Schneider, P . and Wagoner, R . V. ( 1986). Preprint.
Efstathiou , G. and Silk, J . I. ( 1983). Fundamentals Cosmic Phys., 9, 1 .
Einasto , J . , Kaasik, A . and Saar, E. ( 1974). Nature, 250, 309.
Einstein, A. ( 19 17). Sitz. Preuss. Acad. Wiss., 142.
Faber, S . M . ( 1984) . Proceedings First ESO/CERN Conference, ed . G. Setti and
L. van Hove.
Fall, S. M . and Rees, M . J. (1 985). Astrophys. J., 298, 1 8 .
Frenk, C. S., White, S . D . M . , Efstathiou, G. and Davis , M . ( 1985) . Nature, 317, 595.
Friedman, A. ( 1922). Z. Phys., 10, 377 .
Goodman, M . W . and Witten, E . (1985). Phys. Rev., D 31 , 3059.
Gott , J. R. ( 1 98 1). Astrophys. J., 243, 140.
Gott, J. R . , Gunn, J. E . , Schramm, D. N. and Tinsley, B. M. ( 1974). Astrophys. J., 194,
543 .
Grishchuk, L . and Zel'dovich, Y . B . ( 1978). Sov. Astron., 22, 125.
Gunn, J. E., Lee, B . W., Lerche, I., Schramm, D. N. and Steigman, G. ( 1 978).
Astrophys. J., 223, 1015.
Guyot, M . and Zel'dovich, Y . B . ( 1970). Astron. Astrophys. , 9, 227.
Harrison , E. R. ( 1970). Phys. Rev., D l , 2726.
Hartquist , T. W. and Cameron, A . G . W. ( 1977) . Astrophys. Sp. Sci. , 48, 105 .
Hawking , S. W . and Ellis, G . F . R . ( 1968). Astrophys. J., 152, 25.
Hodge, P . W. ( 198 1). Ann . Rev. Astr. Astrophys., 19, 357.
Hogan , C . ( 1986). In Inner Space/Outer Space, ed . E. W. Kolb et al., p . 279. Chicago
University Press : Chicago.
Hogan, C . , Kaiser, N. and Rees, M . J. ( 1982). Phil. Trans. Roy. Soc. , A307, 97.
Hoyle, F. and Tayler, R. J. ( 1964). Nature, 203, 1 108.
Hubble, E . ( 1929) . Proc. Nat. Acad. Sci., 15, 168.
Kaiser, N. (1982). Mon. Not. R. Astron. Soc. , 198, 1033.
Kaiser, N . and Silk, J. I. ( 1986) . Nature, 324, 529 .
Knapp, J . and Kormendy, J . ( 1986) . Dark Matter in the Universe, Proceedings I AU
Symposium 1 1 7 . Reidel : Dordrecht.
Kolb, E. W. and Wolfram, S. ( 1980). Nucl. Phys., B172, 224.
Loh, E. D . and Spillar, E. J . ( 1986) . Astrophys. J. Lett., 307, L l .
Lyubimov, V . A . , Novikov, E. G . , Nozik, V . Z . , Tretyakov, E. F . and Kozek, V . S .
( 1980). Phys. Lett., 394, 266.
McCrea, W. H. and M ilne, E. A. ( 1934). Quart. J. Math . (Oxford), 5, 64.
Madsen, J. and Epstein, R . I. ( 1984) . Astrophys. J., 282, 1 1 .
Marx, G. and Szalay, A . ( 1972). Proc. Neutrino 82 , Technoinform, Budapest, p. 19 1 .
Meszaros, P. ( 1974). Astron. Astrophys., 37, 225 .
498 M. J. Rees
Ostriker, J. P., Peebles, P. J. E. and Yahil, A. ( 1 974) . Astrophys. J. Lett., 193, L l .

Paczynski, B. ( 1986a). Astrophys. J . , 301 , 503 .
Paczynski, B. ( 1986b). Astrophys. J., 304, 1 .
Partridge, R. B. ( 1986). Proceedings /AU Symposium o n Cosmology (in press).
Peebles, P . J . E. ( 1966). Astrophys. J., 146, 542.
Peebles, P . J . E. ( 198 1). Astrophys. J., 248 , 885.
Peebles, P. J . E. ( 1982). Astrophys. J., 258, 4 15.
Penzias, A. A . and Wilson, R . W. ( 1965). Astrophys. J., 142, 4 19.
Press, W. H . and Gunn , J . E. ( 1973). Astrophys. J., 185, 397.
Press, W. H. and Vishniac, E. ( 1 980). Astrophys. J., 236, 323 .
Rees, M . J . ( 1983). I n The Very Early Universe, ed. G. W. Gibbons et al., p . 29 .
Rees, M . J. ( 1985). Mon. Not. R. Astron. Soc., 213, 75P.
Rees, M . J . ( 1 986). Proceedings 2nd ESQ/CERN Conference, ed. G . Setti and
L. van Hove.
Rees, M. J. and Ostriker, J . P . ( 1 977) . Mon. Not . R. Astron. Soc., 1 79 , 54 1 .
Rees, M . J . and Sciama, D . W. ( 1968). Nature, 2 1 7 , 5 1 1 .
Refsdal, S . ( 1964). Mon. Not. R. Astron. Soc., 128, 295.
Refsdal, S . ( 1970). Astrophys. J., 1 59, 3 57 .
Rowan-Robinson, M. ( 1985). The Cosmic Distance Scale. Freeman : New York.
Rubin, V. ( 1986). In Highlights of Astronomy, ed. P. Swings. Reidel : Dordrecht .
Ryle, M . ( 1958). Proc. Roy. Soc., A248, 289.
Sachs, R . K . and Wolfe, A. M. ( 1967). Astrophys. J., 147, 7 3 .
Sikivie, P. ( 1985). Phys. Rev., D32, 2988.
"
Silk, J . I . ( 1982). In Astrophysical Cosmology, ed . H . A. Bruck et al., p . 427. Vatican
Publications.
Silk, J. I. ( 1985). Astrophys. J., 297, 1 .
Silk , J . I . , Olive, K . I . and Srednicki, M . ( 1985). Phys. Rev. Lett., 55, 257.
Smith, P . F. ( 1986) . Proceedings 2nd ESQ/CERN Conference, ed. G . Setti and
L. van Hove.
Spergel, D. N. and Press, W. H. ( 1985). Astrophys. J., 294, 663 .
Steigman, S., Sarazin, C. L . , Quintana, H. and Faulkner, J. ( 1978). Astron. J. , 83, 1050.
Tremaine, S. and Gunn, J. E. ( 1979). Phys. Rev. Lett. , 42 , 407.
Uson, J. and Wilkinson, D . T. ( 1984). Nature, 312, 427.
Wagoner, R. V., Fowler, W. A . and Hoyle, F. ( 1 967) . Astrophys. J., 148, 3.
Wilkinson , D. ( 1986). Phil. Trans. R. Soc., A320, 595.
Witten, E. ( 1984). Phys. Rev., D30, 272 .
Yang, J . , Turner, M . S . , Steigman, G ., Schramm, D . N . and Olive, K . A. ( 1984).
Astrophys. J., 281 , 493 .
Young, P . J . ( 198 1). Astrophys. J., 244, 756.
Zel'dovich, Y . B. ( 1972). Mon. Not. R. Astron. Soc. , 160 , l P .
Zwicky, F. ( 1933). Helv. Phys. Acta, 6, 1 10.
11
Gravitational interactions of cosmic strings
A L EXAN D E R V I L EN K I N
1 1. 1 Introduction
The origin of structure in the universe is one of the great cosmological
mysteries. Newton thought that it could not be explained by natu ral causes
and attributed it to God . Now, 300 years after the publication of Principia,
the problem is still unresolved , and possible ways to approach it have
started to emerge only in the last several years. One of the possibilities is
related to cosmic strings, which could arise as a random network of line-like
defects at a phase transition in the early Universe. In this scenario, massive
closed loops of string serve as seeds for the formation of galaxies and clusters
of galaxies. While matter is being accreted onto the loops, they oscillate
violently, lose their energy by gravitational radiation , shrink and disappear.
Apart from their possible role in galaxy formation , strings are fascinating
objects in their own right. The physical properties of strings are very
different from those of more familiar systems and can give rise to a rich
variety of unusual physical phenomena. In particular, if strings exist, they
can produce a number of characteristic observational effects detectable with
existing astronomical instruments.
The physical properties, evolution and cosmological consequences of
strings have been studied extensively for several years, but the subject is still
rapidly expanding. For an up-to-date guide to the literature the reader is
referred to Vilenkin ( 1985) and Preskill ( 1985). A review of all aspects of
cosmic strings would require writing a book, and here I have chosen just one
aspect which I think is the most appropriate one for a volume dedicated to
Newton's Principia.
This article reviews the gravitational properties of cosmic strings. Some of
these properties have been used to analyse various cosmological effects of
strings, but here my emphasis will be on the basic physics. Cosmological and
500 A . Vilenkin
astrophysical applications will be mentioned only occasionally. The article

can be roughly divided into three parts. The first part, including Sections
1 1 .2-1 1.6, reviews the dynamics of strings in flat spacetime: After a brief
summary of the relevant physical properties of strings, I derive the string
equations of motion and discu ss their solutions describing oscillating closed
loops and waves on infinite strings. In the second part (Sections 1 1.7- 1 1 . 10) I
give some exact solutions of Einstein's equations for static straight strings
and analyze their physical properties. In the third part (Sections
1 1. 1 1-1 1 . 14) the gravitational field of oscillating strings is studied in the
linear perturbation theory. The string configurations found in the first part
of the paper are used as sources of the linearized Einstein equations. The
gravitational field produced by oscillating closed l oops and the gravitational
radiation from loops are discussed in detail .
Throughout the paper I use the system of units in which h c 1 . The
= =
metric signature is ( + , - , - , - ). Greek indices take values from 0 to 3 and

Latin indices take values 1 to 3, or, where it is specifically indicated, only the
values 1 and 2. Notation (x 0 , x1 , x2, x3) and (t, x, y, z) for spacetime
coordinates is used interchangeably.
1 1.2 Some properties of strings

Strings arise in gauge theories with spontaneously broken symmetries. If the
symmetry groups satisfy certain topological conditions, the field equations
of the theory have stable string-like solutions with a unit flux of a massive
gauge field running along each string. A close condensed-matter analogue of
strings is a quantized tube of magnetic flux in superconductors.
Another type of string, called a global string, is similar to a vortex line in
liquid helium. It arises as a result of a global symmetry breaking and does
not carry a gauge magnetic flux. Such strings have long-range interactions
due to their coupling to a massless Goldstone boson. This coupling is much
stronger than thei� coupling to gravity, and so gravitational interactions do
not play a significant role in the dynamics of global strings (Davis, 1985). In
this paper we shall concentrate on gauge-symmetry strings, for which
gravity is the dominant long-range interaction.
In some gauge models, strings do not have end s : they are either infinite
or closed. This is the most interesting type of string for cosmological
applications. In other models strings can end at magnetic monopoles. In this
case the magnetic flux of the monopoles is confined in the string.
Some properties of strings can be deduced just from dimensionality and
Lorentz invariance arguments. Strings are typically characterized by a
Gravitational interactions of cosmic strings 501
single dimensional parameter Y/ - the energy scale of symmetry breaking.

Then , from dimensionality, the thickness of the string is c5 "" 1J -.1 and the
mass per unit length of string is
(2 . 1)
0
For a grand unification scale YJ - 101 6 GeV we get c5 - 10 - 3 cm and µ -
10 22 g/cm. Strings of cosmological interest have sizes much greater than
their thickness and can be treated as one-dimensional objects.
Consider now a static straight string lying along the z-axis. It is described
by a solution of Lorentz-invariant field equations which is independent of z
and t, and should therefore be invariant under Lorentz boosts along the z
axis. Hence the rest frame of the string is defined only up to longitudinal
Lorentz boosts, and only transverse motion of the string has physical
meaning. The physical characteristics of the string , such as the energy
momentum tensor T�, should have the same invariance. This implies that
T8 = T� (2 .2)
with all other components equal to zero, except perhaps T} with i ,j = 1, 2.
Using the conservation law TL = O (in flat spacetime) it can be shown that
the latter components vanish when averaged over the cross-section of the
string
f f
O = T�l , J.xk d2 x = - r�I aJ.xk d2x,
f
= - n d2 x (i,j, k = 1 , 2). (2 .3)
Then , neglecting the width of the string, we can write
T� = µ c5(x) c5(y) diag( l , 0, 0, 1 ). (2 .4)
We see that the tension along the string is equal to the linear mass density.
1 1.3 Nambu action

A curved string oscillates under the influence of tension. Its world history
corresponds to a two-dimensional surface in spacetime
x µ = xµ (( 0 , ( 1 ) , (3 . 1)
where ( 0 and ' 1 are, respectively, a time-like and a space-like parameter on
this surface. (Here again we neglect the string thickness, c5, assuming that it is
much smaller than the curvature radius of the string.) The action functional
for a string should satisfy the following requirements :
502 A. Vilenkin
(i) It should be invariant under general coordinate transformations,

xµ --+ xµ' (x v ) (3 .2)
and under arbitrary reparametrization of the surface,
(a --+ (a' (( b). (3 .3)
(ii) It should be local, since there is no long-range interaction between
different parts of the string . t The general form of the action is then
s = f ff( - y )''' d ' " (3 .4)
where the Lagrangian ff' is invariant under (3 .2) and (3 .3), y is the
determinant of the metric tensor of the su rface,
ax µ a x v
Ya b = 9 µ v a(a a( b , (3 .5)
Y = .xzx,z - (x . x') 2 (3 .6)

and gµv is the four-dimensional metric. Here, dots and primes stand for
derivatives with respect to ( and ( 1 , respectively, x 2 = x µxµ , etc.
0
The 'building blocks' for the Lagrangian ff' are the string tension µ and the
geometric qu a ntities, such as the intrinsic and extrinsic curvature of the
surface (3 . 1) and their covariant derivatives. Note that the 4-velocity uµ is
not among the building blocks; the reason is that the local rest frame of the
string, and therefore the 4-velocity, is defined only up to longitudinal
Lorentz boosts.
The dimension of ff' is mass squared, and we can write
ff' = - µ + r.xK + {3K 2/µ + · , · · (3 .7)
where K stands for the curvature (with indices suppressed) and a , {3 are
numerical coefficients. For a string with a typical curvature radius R , we
have K ,...,, R - 2 4:. µ (since the string thickness is b ,...,, µ - 1 ' 2 and we assumed
that R � b). Hence, the curvature-dependent terms in (3.7) can be neglected
and we obtain
s =
f
- µ ( - rl ' '' d ' " (3.8)
This is the Nambu action for a string (see, for example, Scherk, 1975). Up to
an overall factor, it is given by the surface area traversed by the string in
t Note that the second requirement is not satisfied for global strings. The long-range
force in this case is due to the interaction of strings with Goldstone bosons; the
corresponding action functional is derived in Vilenkin and Vachaspati ( 1986).
spacetime. Note that (3.8) is similar to the action for a relativistic particle,
f
S = - m ds , (3 .9)
which is proportional to the length of the particle's world line.
Varying the action (3 .8) with respect to x µ ((a) we obtain the equations of
motion for a string :
a
ac o { ( - y ) - 1 / 2 [(.X . x')x' µ - x' 2_xµ] }
a
+ ac 1 {( - y) - 1 1 2 [(.X . x')xµ - x 2 x' µ] } = 0. (3. 10)
The energy-momentum tensor can be found by varying S with respect to gµ v
(Turok , 1984; Vachaspati , 1986) :
f
Tµ' = µ d ' (( - y) - 1 1 2 b( •l (xµ - xµ ((" ))
x
{x' 2 xµ_x v + x 2 x' µx' v - (x · x')(xµx' v + x' µx v )} . (3 . 1 1)
It is easily verified that for a straight string in flat spacetime lying along the z
axis , t = (0, z = (1 , x = y = O, and eq. (3 . 1 1) reduced to eq. (2.3).
1 1.4 Gauge conditions

Since the Nambu action is invariant under transformations (3.3), we are free
to choose any parametrization of the world surface. This can be done
explicitly, for example,
(4. 1)
or by imposing two gauge conditions. A very convenient choice of gauge for
a string in flat spacetime is (Scherk, 1975)
x · x' = O, x 2 + x' 2 = 0. (4.2)
The string equation of motion in this gauge is simply the wave equation ,
x1' - x"µ = O. (4.3)
We note further that t = x0 = (0 solves eq . (4.3). With this choice the string
trajectory can be written as
x(C. t), (4.4)
where C = (1 is the space-like parameter on the string, and eqs. (4.2), (4.3)
take the form
x•• · x = 0.. ; x: 1 + x+1 2 = 1 ;
+/
(4.5)
.x - x" = O. (4.6)
504 A . Vilenkin
The energy-momentum tensor is also greatly simplified in this gauge :
I ·
T"'(F, t) = µ d\ (x• x -x'•x") b (3l(f - X(\, t)). (4. 7)
The total energy of the string is given by
J
E = rg d 3x = µ d\. J (4.8)
In the following two sections we shall find some solutions of the string
equations of motion in flat spacetime .
1 1.5 Oscillating loops

The general solution of eq. (4.6) is
x(C t) = -!-[a(( - t) + "b(( + t )J , (5. 1)
and eqs. (4.5) give the following constraints for the otherwise arbitrary
functions a and b (Kibble and Turok , 1982) :
a' 2 = b' 2 = 1 . (5 .2)
The motion of a closed loop of string in its center-of-mass frame is described
by a solution of the form (5. 1) where a(O and b (O are periodic functions,
a(( + L) = a(O; b (( + L) = b ((). (5.3)
From eq. (4.8) we find
L = M/µ, (5 .4)
where M is the mass of the loop. It is clear from eq. (5. 1) that the motion of
the loop must also be periodic in time. The period is T = L/2 rather than
,
T = L, since
x(( + L/2, t + L/2) = x((, t). (5.5)
In fact, the period can be smaller than L/2 for some special loop trajectories.
An interesting property of the loop solutions is that the string typically
reaches the velocity of light at some points at certain moments during its
period (Turok , 1984). From eq. (5. 1) we have
i 2 (C t) = i[a'(( - t) - "b' K + t)] 2 • (5.6)
Now, it follows from (5 .2) and (5.3) that the vector functions a'(O and - b '(O
describe closed curves on a unit sphere as ( runs from 0 to L. These functions
should satisfy
(5.7 )
and are otherwise arbitrary. If the two curves intersect, a'((a) = - b'((b), then
eq. (5.6) gives x 2 ( ( , t) = 1 for 2( = (a + (b + nL, 2 t = ( b - Ci + nL, where n is an

integer. It is possible to construct loops which never reach the velocity of
light, but that requires a somewhat contrived choice of the functio n s a(() and
b(().
To study the behavior of the string near the point of luminal motion , it is
convenient to choose the origin of spacetime coordinates and the
parametrization of the string so that the luminal point is at ( = t = 0 and
x = 0. Expanding the functions a and b near ( = 0,
a(O = a�( + 1a�(2 + i-a b3 > (3 + · · · ,
b(O = b�' + 1b�(2 + i-b b3 > (3 + · · · , (5.8)
and using eq. (5 .2) we obtain
+/
a o = - 7D 1o
1 a� 1 = 17J� I = 1 , (5.9)
_ 7,
a.,o · a·,,o - .
Do D o7" - 0
o = -b ,,z
a.,,o z + a·'o · a·(3) o +7' Do = 0 , · · · ·
Do . 7(3)
The first of these relations just expresses the fact that the curves li'(O and
- b'(O intersect at ( = t = 0. From eqs. (5.6), (5.8) and (5.9) we find
i1(C t) = 1 - -i [(( - t)a� + (( + t)b�J 2 + · · · . (5 . 10)
The shape of the string at t = 0 is given by
x(C O) = i (a� + b� K 2 + / (a b3> + bb3 > )( � + · · · . (5. 1 1)
2
It is easily seen that for (a� + b�) # 0 the string momentarily develops a cusp,
as shown in Fig. 1 1 . 1 , and it follows from eq. (5.9) that the direction of the
cusp (a� + b�) is orthogonal to that of the luminal velocity ( - a�). With an
Fig. 1 1 . 1 . The string momentarily develops a cusp at the point of luminal

motion .
x
506 A . Vilenkin
appropriate choice of axes (Fig. 1 1 . 1 ) the shape· of the string near a cusp is
given by y ex: x 21 3 •
The rms string velocity in a loop can be defined as v rms ( ( v 2 ) ) 11 2, where
=
(v 2 ) = 1 r dt � I d(.i2(" t)
= 1 -1( ci '(( - t) . b' (( + t)) (5. 12)
and I have used eq. (5.6). Using the identity
:1 J J' · b d ( = J ( - J" · b + J' · b') d (
=2 J a' · b' d( (5. 13)
and the fact that an average over period of a time derivative is equal to zero,
we see that the last term in (5. 12) vanishes, and thus
( v 2 ) = 0.5. (5. 14)
To give a specific example of loop trajectories, we can choose a and b to
describe circular motion in planes at an angle </> to one another :
ci((} 'ix - 1 (e 1 sin ix( + e 3 COS IX(},
b(() = p - 1 [(e 1 cos <J> + e sin </>) sin /J( + e 3 cos /J(] . (5. 1 5)
2
Here, ei is a unit vector along the xi-axis, ix = 2nm/L , p = 2nn/L, m and n are
relatively prime integers (otherwise the parameter ( traverses the loop more
than once between 0 and L and can be redefined to make m and n relatively
prime). It is easily checked that the period of the loop (5. 1 5) is T = L/2n. The
family of solutions (5 . 1 5) was found by Burden ( 1985) . Solutions with
(m, n) = ( 1 , 1) as well as other families of solutions had been studied earlier by
Kibble and Turok (1982) and Turok ( 1984).
Eq . (5. 1 5) with m = n = 1 describes an elliptical loop which rotates and
turns into a double line; at that moment the ends of the line are moving at
the speed of light. Then the loop returns to the elliptical shape and goes
through the cycle again. The degenerate cases </> = 0 and </> = n correspond to
an oscillating circular loop and to a rotating double line, respectively. Loops
with (m, n) # ( 1 , 1) never collapse to a double line; in fact, loops with m = 1 ,
n # 1 never self-intersect. Several snapshots of a loop with m 1 , n = 2, =
</> = n/2 are shown in Fig. 1 1 .2.
1 1.6 Waves on infinite strings

To illustrate another type of string motion, it is convenient to use the gauge
(4. 1). Then the string trajectory is described by two functions, x(z, t) and
y(z, t), and it is easily verified that
x =f(z + t), y = g(z + t) (6. 1)
is a solution of eq . (3 . 10) for arbitrary functions f and g. These solutions
describe waves of arbitrary shape propagating along the string with the
velocity of light. Note that a superposition of waves travelling in opposite
directions is not a solution, since eq. (3 . 10) is nonlinear.
1 1.7 Gravitational field of a straight string

In this section we shall find a solution of Einstein's equations describing the
gravitational field of a static straight string lying along the z-axis. Such a
string is left unchanged by time translations , spatial translations in the z
direction , rotations around the z-axis and Lorentz boosts in the z-direction.
Assuming that the metric has the same symmetries, we can write it as
d s 2 = A 2 (p )(dt 2 - d z 2 ) - dp 2 - B 2 (p ) d¢ 2 . (7. 1)
The energy-momentum tensor of the string vanishes exponentially at p � <5 ,
where <5 is the string thickness, and s o outside the string core the metric is
accurately described by the vacuum Einstein equations. Kasner ( 192 1) has
shown that the general vacuum solution depending on one spatial variable
has the form (up to a coordinate transformation)
ds 2 = p 2p1 dt 2 - d p 2 p 2P2 d¢ 2 p 2PJ d z 2 ,
_ _
(7 .2)
P 1 + P 2 + p 3 = P f + P � + P� = 1 . (7 .3)
The metrics (7 . 1) and (7 .2) are consistent with one another only if p 1 p 3 • =
The only solutions of eq. (7 .3) with p 1 = p 3 are

(7.4)
P 1 = p 3 = i , Pi = - ! . (7.5)
In the solution (7.5), the length of a circle, p = const, decreases as p - 1/ 3 when
Fig. 1 1 .2 . The xz-projections of the loop, e.g. (5. 1 5), with m= 1 , n = 2,

</> = n/2 at four moments during its period .
508 A . Vilenkin
p --+ oo . It is unlikely that this metric has anything to do with strings which
could have formed in our Universe, and we dismiss the case (7 .5) as
unphysical . The remaining solution (7 .4) is just the metric of flat space in
cylindrical coordinates,
ds 2 = d t 2 - dp 2 - p 2 d</> 2 - d z 2 • (7 .6)
Thu s, we have reached a surprising conclusion that the spacetime outside a
straight string is flat. We shall see, however, that it is only locally flat and
that globally it is not equivalent to Minkowski space (Vilenkin, 198 la).
To represent the interior of the string , we shall first assume a simple
energy-momentu m tensor of the form (Linet, 198 5 ; Gott, 1985; Hiscock,
1985)
T� = a(p) diag( l , 0, 0, 1). (7.7)
The general case will be discussed later. The form (7.7) is clearly con sistent
with all the symmetries of the string. In the limit of negligible string
thickness, a(p) is a b-function and (7.7) reduces to (2 .4) . Substituting (7 . 1)
and (7. 7) in Einstein's equations, it is easily shown that A(p) = con st . By a
suitable rescaling of t and z we can set A = 1 ; then
ds 2 = dt 2 - dz 2 - dp 2 - B 2 (p) d</> 2 (7 .8)
and Einstein 's equations reduce to a single equation for the function B(p) :
B"/B = - 8nGa. (7 .9)
The metric (7 .8) is nonsingular on the axis p = 0 only if
B(O) = O, B'(O) = 1 . (7 . 10)
The mass per unit length of string is
2n
('xi dp [ d</Ja(< 2 >g) 1 1 2 = - [ 1 - B'( oc )] ,
1
Jo Jo
µ= (7. 1 1)
4G
where < 2lgii is the metric on the surface (t, z) = const and < 2lg = B 2 is its
determinant. At large distances from the string (p � b) , a --+ 0, B'(p) -+ B'( oo )
and we obtain
ds 2 = dt 2 - dz 2 - dp 2 - ( 1 - 4Gµ) 2p 2 d <f> 2• (7 . 12)
A coordinate transformation
( l - 4Gµ)</J --+ </> (7 . 13)
brings the metric (7. 14) to a locally Minkowskian form (7.6), but then the
angle </> varies in the range
0 < </> < ( 1 - 4Gµ)2n . (7 . 14)
Thus, the effect of the string is to introduce an azimuthal 'deficit angle '
<5 = 8 n Gµ , (7. 1 5)
with the result that a surface of constant t and z has the geometi::y of a cone
rather than that of a plane. The point of the cone is smoothed on a scale
P "' <5. In the limit of an infinitely thin string the cone has a sharp point ; the
· corresponding spacetime is called conical.
The dimensionless parameter Gµ plays an important role in the physics of
cosmic strings. Its magnitude can be estimated using eq. (2 . 1),
Gµ - (17/mp) 2 � 1 , (7. 16)
where 17 is the symmetry breaking scale of strings, mP is the Planck mass and
it is assumed that 17 � mP . The string scenario of galaxy formation requires
Gµ ....., 10 - 6 , which corresponds to 11 "" 10 1 6 GeV .
1 1.8 Gravitational field of a straight string (continued)
We derived eq. (7 . 15) for the conical deficit angle assuming a simple form of
the energy-momentum tensor (7.7). In the general case, the components T�
and T$ do not vanish inside the string and eq. (7 . 1 5) has to be modified. We
shall see, however, that this equation holds in the cosmologically interesting
case of Gµ � 1 (the reason , basically, is that for Gµ � 1 the metric inside the
string is approximately flat and that transverse tensions average out to zero
in flat spacetime (see eq. (2 .3)).
We shall first express the deficit angle in terms of the curvature tensor in
the interior of the string (Ford and Vilenkin, 198 1); Let S denote a surface
(t, z) = const and let <2lg ij be the metric on S. The scalar curvature of S is
<2>R = 2B /B.
- " (8 . 1)
Integrating <2>R over the surface and using eq. (7. 12) we obtain the following
expression for the deficit angle :
f
.5 = � 1 2' R ( ' 2 'g) ' f 2 d 2 x. (8 .2)
This is a version of the Gauss-Bonnet theorem.

The curvature of the surface S can be expressed in terms of the Riemann
tensor <4l Rear of the four-dimensional spacetime. First consider a 3-surface
t = con st. Its Riemann tensor < 3 > R)k1 is expressible in terms of <4l Rear and the
extrinsic curvature K ij by means of the Gauss-Codazzi equations. However,
for the 3-surface t = const, the extrinsic curvature vanishes and < 3 > R)ki =
<4>R)k i · Similarly, one obtains that the Riemann tensor of the 2-surface S is
i. = <4> RJkl•
<2> RJkl i. Hence '
(8 .3)
510 A . Vilenkin
This result, combined with eq. (8.2), enables us to find the deficit angle in
terms of the Riemann tensor of the four-dimensional spacetime.
In general, it does not seem to be possible to express o in terrris of the four
dimensional Ricci tensor, and hence of the energy-momentum tensor of the
string, T,w However, in the case that the gravitational field is sufficiently
weak that the linearized theory may be applied, such an expression can be
given. Let the metric be
gµv = 1Jµv + hµv • (8 .4)
where 1'/ µv = diag( 1, - 1, - 1, - 1) and hµv � 1 . With the gauge condition
a\l(h; - !o;h) = O (8 .5)
the linearized field equations become
O hµ v = - 1 6 nG Sµv • (8 6)
.
where D = a; - v 2 , the indices are raised and lowered using the flat metric
t'lµv and
S µv = Tµv - 111µv T. (8.7)
The linearized Riemann tensor is
( 4) Ra.µpv = !(h a.v,µp + hµp, a.v - hµv, a.p - ha.p, µv ) · (8 .8)
For a metric independent of t and z,
< 2 > R - 2< 4) R 1 2 1 2 - 2h 1 2 , 1 2 - h 1 1 , 22 - h 22 , 1 1 • (8 .9)
Using (8 .5) and (8 .6), this can be rewritten as
< 2 > R = - 1 6 nG( T1 1 + T2 2 - ! T). (8 . 10)
Substituting this in (8 .2) and using (2.2), (2.3), we obtain
S = SnG f T8 d 2 x = SnGµ . (8 . 1 1)
The use of the linear perturbation theory is justified for Gµ � 1 , which is the
case for strings of cosmological interest.
A non-perturbative analysis of Einstein's equations with a realistic
energy-momentum tensor of a string has been given by Garfinkle ( 1985).
The gravitational field of global strings, for which the energy-momentum
tensor falls off only as an inverse square of the distance from the string , has
been discussed by Aryal and Everett ( 1986). I should mention also that
spacetimes with conical singularities had been studied long before their
relevance to cosmic strings was recognized . See, for example, Bach and Weyl
( 1922), Sokolov and Starobinsky ( 1977) and Israel ( 1977).
Gravitational interactions of cosmic strings 51 1
1 1.9 Physics in conical space

We have found that, neglecting the string thickness , the gravitational field of
a straight string is given by the metric (7 . 12). It describes a conical space,
which is just a flat space with a wedge of angular size b = 8nGµ removed and
the two faces of the wedge identified . The geodesics in this space are straight
lines, and it is clear that a test particle initially at rest relative to the string
will remain at rest and will not experience any gravitational force.
Although the metric (7. 12) is locally flat, its global structure is different
from that of M inkowski space. The most striking effect of this difference is
the formation of double images of objects located behind the string.
(Vilenkin , 198 l a , 1984 ; Gott , 1985). This is illustrated in Fig. 1 1 .3. The light
rays from a quasar intersect behind the string and the observer sees two
images of the quasar. If l and d are the distances from the string to the quasar
and to the ob server, respectively, and f) is the angle between the string and
the line of sight, then it is easily shown that the angular separation between
the images is (for Gµ � 1).
b</> 0 = 8nGµl(d + l) - 1 sin fJ. (9. 1)
Gravitational lensing by a string is a classical analogue of the Aharonov
Bohm effect. Spacetime curvature is confined to the string core, but its effect
is 'felt' by the photons propagating in flat spacetime region around it. As in
the Aharonov-Bohm case, a Minkowskian coordinate system can be chosen
in any region on one side of the string, but such a system does not exist in a
region surrounding the string.
The subscript of £5¢ 0 in eq . (9. 1) refers to the fact that the string was
assumed to be at rest with respect to the observer. Cosmological strings are
expected to move at relativistic speeds, and so eq . (9. 1) has to be generalized
to the case of a moving string. In the general case, let s and v be the direction
and the velocity of the string in the observer's frame S. Since only transverse
motion of strings is observable, we can take v _l s. Let k and k' be the four-
Fig. 1 1 . 3 . Light rays emitted by the quasar intersect behind the string and
the observer sees two images of the same quasar.
Observer
"-,,,
�
Observer
512 A . Vilenkin
dimensional wave vectors of the light waves corresponding to the two

images and consider the invariant
k · k' = ww' [ l - cos(.5 ¢)] :::::: !ww' (.5 ¢) 2 . (9.2)
In the rest frame of the string� S 0 , the two frequencies are equal , w = w' = w 0 .
In observer's frame, S, w and w' are slightly different because of the
difference in the directions of k and k' . However, the difference is ,.., .5 ¢ and
can be neglected to the lowest order in Gµ . Then rob¢ = w 0 b¢ 0 . The
frequencies w and w 0 are related by w = y( l - ii · v)w 0 , where n is the direction
from the observer to the quasar and y = ( l - v 2 ) - 1 ' 2 . Hence,
.5¢ = y - 1 ( l - n · v) - 1 .5 ¢ 0 • (9 . 3)
Depending on the relative direction of v and ii , the angular separation takes
a value in the range
K l /2 b cP o < b cP < K - 1/2 b¢ o , (9.4)
where K = ( 1 - v)/( l + v).
For Gµ ,...., 10 - 6 , the typical angular separation of images is of the order of a
few seconds of arc. Several double quasars with such angular separations are
known , and it has been argued that some of them may be due to strings
(Gott, 198 5 ; Vilenkin, 1984; Paczynski, 1 986).
Another interesting effect arises because the relative velocity of two
objects in a conical space is double-valued . The answer depends on which
way one parallel transports the velocities around the string. If velocities are
measured using direct light signals, then the relative velocity experiences a
discontinuous change when the string passes between the two objects. In a
cosmological setting , this results in a steplike discontinuity of the microwave
background temperature along the string . The velocity of the observer
relative to the surface of last scattering has different values in front of and
behind a moving string ; the corresponding Doppler shift is responsible for
the temperature discontinuity.
Let us now describe this effect quantitatively (Kaiser and Stebbins, 1 984;
Vachaspati, 1986; Vilenkin, 1986). In the rest frame of the string, S 0 , the
observer moves by the string with velocity - v. If he is at rest with respect to
the sources of radiation on one side of the string, then the velocity of the
sources on the other side of the string is -v + u 0 , where u 0 = 8nGµv x s. The
corresponding velocity difference in the observer's frame S is u = yu 0 • (Here,
I have expanded u to the lowest order in Gµ and used the fact u 0 · v = O.) The
temperature discontinuity can be found from the Doppler formula,
.5 T/T = u · n ,
.5 T /T = 8nGµyn (v x s).
· (9 .5)
The existence of strings is consistent with the observed isotropy of the

cosmic microwave background (bT/T < 10 - 4 on large scales) only if
Gµ < lO - s .
Smith ( 1986) has studied the effects of a conical singularity on classical
and quantum fields. He has found , in particular, that although a straight
string exerts no gravitational force on a particle outside it , the particle's own
gravitational field is distorted by the string and results in a classical
attractive self-force proportional to the particle mass squared :
F = inGµ Gm 2/r 2 • (9.6)
This result was also obtained by Linet ( 1 986).
1 1.10 Other metrics with straight strings

Solutions of Einstein's equations with conical singularities describing
straight strings can easily be constructed (Aryal , Ford and Vilenkin , 1986).
One needs only a spacetime with a symmetry axis. If one then cuts a wedge
out of this space, which is done by requiring the azimuthal angle around the
axis to run over the range ( 7 . 14) or by multiplying g41q, by ( l - 4Gµ) 2 , one has
a space with a string along the axis.
( 1 - ---;-2m) dt 2 - ( 1 - ---;-2m) -
For example, the metrics
i
ds 2 = dr 2 - r 2 d .i.nt 2 , ( 10. 1)
d s 2 = d t 2 - a 2 (t)[( 1 - kr 2 ) - 1 d r 2 + r 2 d Q 2] ( 10.2)
with
d Q 2 = d8 2 + ( 1 - 4Gµ) 2 sin 2 e d</> 2 ( 10.3)
describe a black hole with a string passing through it, and a Robertson
Walker universe with a string, respectively.
Some well-known solutions of Einstein's equations can be re-interpreted
in terms of cosmic strings. For example, the solution found many years ago
by Bach and Weyl ( 1922) describes, with an appropriate choice of
integration constants, a pair of black holes held apart by cosmic strings
extending to infinity in opposite directions. If one of the black holes is
removed to a very large, but not infinite, distance, one obtains a solution
representing a black hole suspended by a string in a weak uniform
gravitational field, g. To linear o rder in g,
d s 2 ( 1 - 2m/r)( 1 + 2gz) dt 2 + ( 1 - 2gz)[( 1 - 2m/r) - 1 dr 2
=
+ r 2 d8 2 + b 2 r 2 sin 2 e d</> 2] ' ( 10.4)

514 A . Vilenkin
where z= (r - m) cos 8 ,b=

1 for z < - m and b=
1 - 4mg for z > m. The string
tension µ = mg exactly balances the weight of the black hole.
1 1. 1 1 Gravitational field of oscillating loops

The most striking feature of the gravitational field of a straight string is that
there is no Newtonian force. Its origin can be understood using the linear
perturbation theory (8 .4}-(8 .7). The source of the Newtonian potential ,
<P = !h 00 , is
- 2( T 0 - Ti)
i .
- 8 00 _
S oo _ 1 o ( 1 1 . 1)
For the energy-momentum tensor of a string, eq. (4.7),
sg(r. t) = µ Jd(i 2 J(3l(f - X (C t)) . ( 1 1 .2)

In a rest frame of a straight string, sg =0 and the gravitational potential
vanishes. A string can be static only if it is straight, and so curved strings
produce non-zero gravitational potentials. In this section we shall analyze
the gravitational field produced by oscillating closed loops.
The retarded solution of the wave equation (8.6) can be written as
-r' I sµv(r, t- l r -r' j) .

hµv(r , t) = - 4G Ilrd3r' ( 1 1 .3)
Using eqs. (4.7) and (8 .7) and integrating over r' we obtain (Turok, 1984;
Vachaspati , 1986)
hµv(r, t) = - 4Gµ Il r -x((,Fµv((;r ) d( .
r)l(l - 8 . x((, r))
' ( 1 1 .4)
where
•
= r-x(Cr
t: l r - .X(( , r)) I ' ( 1 1 .5)
Fµv= xµxv - x'µx' v +11µvx'ax'a ( 1 1 .6)
and the retarded time r is defined by
r=t- l r- .X((,r )I . ( 1 1 . 7)
The motion of non-relativistic test particles is determined mainly by the
time-averaged field of the loop,
2 fL/2
( hµv(r) ) =
L Jo
dthµv(r, t). ( 1 1 .8)
=
The period of the oscillating component of the field, T L/2, is much shorter
than the time it takes for the particle to traverse a distance - L. The effect of
the oscillating component on the particle trajectory at a distance ;S L from
the loop i s to introduce a small perturbation of the position aT 2 -6.x ,....,

Gµ L. Here, a - GM//J is the typical acceleration and M = µL is the mass of
the lo �p. The acceleration can be enhanced by the 'beaming factor'
(1 - l ·x)-1 in eq . ( l l .4) (Turok , 1984). This factor becomes infinite along the
beams extending from the cusps where = 1 in the direction of i Because of
lj l
the retardation , the divergence occurs at different moments for different
points along the beam. If a particle passes near the beam at the right time,
the effect can be very large, but it affects only a rather special class of particle
trajectories. Here, we shall first discuss the average field ( 1 1 .8) and then
consider what happens at the beams. The gravitational radiation from loops
is discussed in the next section.
U sing eqs. (1 1 .4), ( 1 1 .8 ) and the relation
dt=d-r[l -e · x(C-r)J, .
( 1 1 .9)
f. f.
we obtain for the average field (Turok, 1983)
BGµ L/2 dt L d( pµv(r -r)
( hµv(r) ) = - -
L o o lr -x(( , -r) I
1o ' • •
( 1 1 . 10)
•
This result is easily interpreted : the time-averaged field of a loop is equal to

that of the surface traced out by its motion with a surface energy
momentum tensor
uµv = Ll l x' (xµxv -x'µx'v).

2µ
( 1 1 . 1 1)
il I
Here, I have used the expression for the surface element, dS = i i I lx' I d( d-r.
The mass per unit area of the surface is given by u 00 = 2µ/L l l x lx'I ·
We note that becomes singular at the cusps, where l�I = 1 and
uµv = 0. lx'I
However, the singularity is integrable and the total mass of the shell is equal
to the mass of the loop, M :
fu00 dS = L IoL/2 dt IoL d(=µL =M.
2µ
J J ( 1 1 . 12)
We also note an interesting possibility that the peaks of uµv

near the cusps
can be directly observable as overd�nsity regions in the distribution of
matter accreted by the loops.
At a large distance from the loop r "t> L, eq. ( 1 1 . 10) gives
( 1 1 . 13)
where
( 1 1 . 14)
516 A . Vilenkin
The integration in ( 1 1 . 14) is easily performed . For example,

A 00 = 4GµL(x 2) = 2GM , ( 1 1 . 15)
where (i2) is defined by eq. (5. 12) and I have used eq . (5. 14). Similarly, we
find
( 1 1 . 16)
Combining eqs. ( 1 1 . 13), ( 1 1 . 1 5) and ( 1 1 . 16), we see that the long-distance
behavior of the average field coincides with that of a Schwarzschild field of
the same mass (Turok , 1984). It should be emphasized that the
Schwarzschild limit is approached only in the average sense. Since the
motion of the loop is relativistic, the oscillating component of hµv describing
gravitational waves has the same order of magnitude as the average field .
We now turn t o the discussion of �he bursts of gravitational field produced
when the beaming factor ( l - e · x) - 1 in eq . ( 1 1 .4) becomes infinite (T.
Vachaspati and A. Vilenkin, 1986, unpublished results). The main
contribution to hµv in the bursts comes from a small vicinity of the cu sp ,
which we take to be at x = t = ( = 0. Taking the slowly varying factors out of
the integral , we can rewrite ( 1 1 .4) as
h" ' (f, t) "' - 4 GµF't,'r - I roo d(( I -Li(" <)) - I , ( 1 1 . 17)
where F�v = pµv(O, 0) and r is a function of r, t and ( implicitly defined by eq .

( 1 1 .7). The integral in ( 1 1 . 17) can be estimated using the expansion (5.8) near
the cusp and the relations (5 .9). Here, we shall calculate hµv only on the
beam, where f. = - a� = b�. With the notation a 0 = l a �l 2/4, Po = l"b �l 2/4, we
obtain for the points on the beam,
3(t - r) + a 0(( - r)3 - P0(( + r)3 � o ; ( 1 1 . 18)
1 - f, x � a0(( - r )2 + Po (( + r )2 ;
· ( 1 1 . 19)
f. x' � a o (C - r )2 - p0(C + r )2.
· ( 1 1 .20)
Introducing a new variable, u = (/r , and using eqs. ( 1 1 . 18)-( 1 1 .20) and the
relation
8-r/8 ( = f. · x' ( l - f. x) - 1
· ( 1 1 .2 1)
we can rewrite ( 1 1 . 17) as
hµv(z, t) � - 16Gµl (a0/P 0 ) F�v z - 1 ( 1 2P�l t - z i) - 1 l 3 , ( 1 1 .22)
where the z-axis i s chosen along the beam and
I (x) = f�00 du [(u + 1) 3 - x ( u - !)Jr 21 3 . ( 1 1 .23)

l(x)
The integrand of u=u0 =(x113+ 1)(x 113 - 1)-1,
i s singular at but the
singularity is integrable and h µv is finite for z - t :;6 0. For a particular value
x= /(1)=2-2133-112B(!,!)=4.6, B(x,y)= l(x)l(y)/l(x+ y)
1,
beta function .
where is the
= t l-113,
The metric ( 1 1 .22) is singular at z t, h µv oc lz -
of the force acting on a test particle is
and the z-component
Fz oc ( z - t) - 13 sign(z - t).
4 (11.24)
The force which is initially repulsive, grows in magnitude, becomes infinite
and instantaneously changes into an infinite attractive force, which then
decreases. (Note, however, that the total momentum transferred to the
particle is finite.) This burst of gravitational field propagates along the beam
at the speed of light. Of course, the linear perturbation theory cannot be
trusted at points where h µv diverges, but we expect eq . ( 11.24)
to apply for
lz - ti not too small. It is possible that nonlinear effects and the back reaction
of the gravitational field on the string near the cusps make hµ v finite at z t.
Eq .( 11.17) gives the most singular components of the gravitational field
=
on the beam . The analysis of other components and of the gravitational field
slightly off the beam is rather complicated and will not be given here.
1 1.12 Gravitational radiation from loops

Gravitational radiation is the main energy-loss mechanism for oscillating
loops of string. The rate of energy loss can be roughly estimated using the
quadrupole radiation formula :
£,...,, GM2L4w6• (12.1)
The typical frequency of oscillation is w ,...,, L- 1 , and we obtain (Vilenkin,
198 1 b)
(12.2)
y
where the numerical coefficient is independent of the loop size, but may
depend on its shape and trajectory. The lifetime of the loop is
T"'M/£,...,, L/yGµ. (12.3)
The quadrupole radiation formula is reliable only for nonrelativistic
sources. The motion of loops is ultrarelativistic near the cusps, and so the
quadrupole formula cannot be used to calculate y in eq . ( 12.2).
The power of
gravitational radiation from a periodic source of period T = to the L/2
lowest order in G,
and without any fu rther assumptions about the source,
518 A . Vilenkin
f
can be found from the following equations (Weinberg, 1972) :
. '°' '°' dPn
P = E = L., Pn = L., dQ - , ( 12.4)
II II
dQ
dP11 Gw; { *
- - -- T µv (wn, )Tµ (w,1 ....
....k v k) - .lj2 yvv (wn , ....k ) j 2 } . ( 12.5)
7t
_
dQ '
Here, dPn/dQ is the radiation power at frequency wn = 2 nn/ T = 4nn/L per
unit solid angle in the direction of k, l"kl = wn and
Tµ '(w., k) = � f:"
f
dt exp(iw,t) d 3 x exp( - ik · X) T"'(X, t) ( 12.6)
is the Fourier transform of the energy-momentum tensor.

This formalism has been applied to simple loop configurations by Turok
( 1984) , Vachaspati and Vilenkin ( 1985) and Burden ( 1985) . For loops
described by eq . (5. 1 5) the angular distribution dPn/dQ can be expressed
analytically in terms of Bessel functions. The total power, Pn , at frequency wn
is then calculated by a straightforward numerical integration . An
asymptotic analysis at large n shows that
pn "-' COnSt n - 4!3, ( 12.7)
indicating that the convergence of the series in ( 12.4) is slow. The best
strategy, then, is to keep calculating Pn until the asymptotic regime of eq .
( 12.7) is reached and then to use that equation to estimate the remainder of
the series. The resulting values of y for several loop configurations are shown
in Fig. 1 1 .4. We see that, somewhat surprisingly, y is typically ,..., 100.
The large value of y is due partly to the large n contributions in eq . ( 12.4).
For 'normal' sources of radiation , Pn decreases exponentially in the large n
limit. The power-law dependence ( 12.7) in the case of loops is due to the
singular behavior of string near the cusps. The angular distribution of the
radiation is also marked by the presence of cusps. Large amounts of
radiation are beamed from the cusps in the direction ofl uminal velocity. The
analysis in Vachaspati and Vilenkin ( 198 5) shows that near the beam
dP '°' dPn
= L., oc e -
1 ( 12.8)
dQ dQ II
'
where e is the angle between the beam and the wave vector k. The radiation
intensity diverges on the beam, but the divergence is integrable and the total
power is finite. The immediate vicinity of the beam, where the linear
perturbation theory breaks down , contributes very little to the total power
( 12.2) .
The gravitational waves emitted by oscillating loops at different epochs
Gravitational interactions of cosmic strings 5 19
add up to a stochastic gravitational wave background with a scal e-invariant

spectrum (Vilenkin , 198 lc ; Hogan and Rees, 1984;· Vachaspati and
Vilenkin , 1985)
( 12.9)
Here, Q9(w) (wfpc) (dp9/dw), p9 is the energy density of gravitational waves
=
and Pc is the critical density. Eq. ( 12 .9) is expected to apply in a wide range of
frequencies ,
( 12. 10)
With Gµ - 10 - 6, as required by the cosmic string scenario of galaxy
formation , eq. ( 12.9) gives Q9(w) - 10 - 1 . A gravitational background of
such intensity should be observable using the millisecond pulsar (Hogan
and Rees, 1984; Witten , 1984).
1 1.13 Gravitational rocket effect

Gravitational radiation from an oscillating loop carries away not only
energy , but also momentum. As a result, the loop can accelerate like a
rocket. The rate of momentum radiation is given by
dP
J
dPn k
= L: dQ. (13. 1)
dt n dQ
Fig. 1 1 .4 . Gravitational radiation power for the family of loops, eq. (5. 1 5),
with m= 1, n = 1 , 3 , 5 for several values of </J.
'Y
200
+ n = 1
• n = 3 •
• n = 5
150
A
1 00
·-
·
·- ·
.///
./ /+
50
+ ...... + ... /+
0 '--��--'-��
0 0.5
520 A. Vilenkin
It vanishes for loops described by eq. (5 . 15) because of their high symmetry.
In general, one expects IPI to be of the same order as E in eq. ( 12.2). A
.
numerical calculation for several asymmetric loops gives (Vachaspati and
Vilenkin , 1985)
( 1 3 .2 )
with '/ p "' 10.
A loop radiating momentum at the rate ( 1 3 .2) will move with acceleration
v = y P Gµ/L. ( 13.3)
If it starts from rest, then by the end of its life it will reach the velocity
V Vr ,....., '/ Ph ,....., 0 . 1 ,
,....., ( 1 3 .4)
where 1' is from eq. ( 12 3) ..
.
In deriving the estimate ( 1 3.4) we assumed that the direction of P does not
change appreciably during the loop's lifetime. This is far from being clear.
For example, one can argue that angular momentum radiation can prevent
the loop from accumulating a large velocity. The angular momentum of a
loop is / ,...., µL2 and the angular momentum radiation rate can be written as
(on dimensional grounds)
( 13.5)
where y 1 is a numerical coefficient. Let us first consider the component of 1
parallel to 1, which changes only the magnitude of 1. If the torque ( 13.5)
causes the loop to rotate as a solid body, then the corresponding angular
acceleration is ff,.._, GµL- 2 . The time it takes the loop to rotate by about one
radian is '1.t {J- 1 12 (Gµ) - 1 12 L, and the velocity accumulated during this
,.._, ,.._,
time with acceleration ( 13.3) is v - (Gµ) 1 ' 2 � 1 . Note, however, that the
assumption that the loop reacts to a torque like a solid is not well justified .
Besides, even if i� does, the rotation axis is not, in general , at right angles to
the direction of P, and then the loop will accelerate along the direction of 1.
Torques perpendicular to 1 tend to change the direction of angular
momentum. However, because of the large value of I, the loop behaves as a
relativistic gyroscope, and the direction of its rotation axis is very stable
(Hogan, �986). According to eq . ( 13.5), this direction changes on a time
scale, l/rl ,....., L/(y 1 Gµ) , comparable with the lifetime of the loop. To
summarize, our qualitative . arguments suggest that loops can accelerate to
velocities v - 0. 1 , at least if P ·l=1= 0. To reach a reliable conclusion, one has to
study the back reaction of gravitational radiation on the loop. Note also
that here we discussed the gravitational rocket effect in flat empty space. The
effects of the cosmological expansion and of the gravitational drag due to
surrounding matter have been studied by Vachaspati and Vilenkin ( 1 985),

Hogan and Rees ( 1984) and Hogan ( 1986) .
1 1. 14 Gravitational field of waves on infinite strings

Waves propagating along infinite strings are described by solutions of the
form (6. 1 ) to the string equations of motion . In this section we shall study the
gravitational field produced by such waves (Vachaspati, 1986). A wave
propagating in the positive z-direction is given by (in the gauge (4. 1))
x =f(z - t) , y = g(z - t) , ( 14. 1)
where f and g are arbitrary functions. The source of the linearized Einstein's
equations (8.6) is found by a straightforward calculation using eq . (3 . 1 1) :
Sµ v = µFµ v (z - t) 6 (x -f) 6 (y -g), ( 14.2)
where
f ' 2 + g ' 2 -! ' -g' ! ' 2 + g ' 2
-! ' 1 0 -! '
Fµv = ( 14.3)
-g' 0 1 - g'
! ' 2 + g ' 2 -! ' -g' !' 2 + g ' 2
Now it is not difficult to verify that
hµv = hFµv(z - t) ( 14.4)
�ith
=h 4Gµ ln [ r0 2 {(x -f) 2 + (y - g) 2 }] ( 14.5)
is a solution of eq . (8 .6) . Indeed , (a;- - a;)hµv 0 because hµv depends on t and
=
z only in the combination (t - z) . On the other hand, (a; - a;) acts only on
the logarithm and gives the 6-function in sµv. r o in eq . ( 14.5) is an arbitrary

constant of integration .
In the case of a static straight string parallel t o the z-axis, all components
of Fµv vanish except F1 1 F 2 2 1, and the solution can be written as
= =
d s 2 dt 2 - dz 2 - ( 1 - h)(dp 2 + p 2 d ¢ 2 ) ,
= ( 14.6)
where p 2 = x 2 + y 2 . A coordinate transformation
( 1 - h)p 2 = ( l - 8Gµ)p ' 2 ( 14.7)
brings it to the form (7 . 12) . The metric ( 14.6) is locally flat, but for a
nontrivial choice of f and g the metric ( 14.4) has a nonvanishing Riemann
tensor.
To elucidate some interesting properties of the gravitational field ( 14.4) ,
let us consider a pulse of length d propagating on a straight string. We can
choose the origin of z so that at any time t the pulse is localized in the range
522 A . Vilenkin
t � z � t + d. ( 14.8)
An unusual feature of the gravitational field of the pulse is that for values of z
outside this range the metric is the same as for a straight string. Hence, the
pulse exerts a gravitational force on a test particle only for a short period
dt - d when the particle is within the slab ( 14.8). Suppose the particle is
initially at rest at a large distance, p, from the string (much greater than the
amplitude of the pulse). Then it can be shown that, after the pulse has passed
by, the particle has a velocity
v= � fr2 + g")
4 µ
dt ( 14.9)
directed towards the string. Note that an analogous situation arises in

electromagnetism when one considers the electric field of an ultra-relativistic
charge. In the limit v --+ 1 , the field is confined to the plane perpendicular to
the direction of motion. The impulse received as this plane passes through a
test charge is also inversely proportional to the distance from the source.
Acknowledgements
I am grateful to Tanmay Vachaspati for many helpful discussions and to
Bruce Allen for his useful comments on the manuscript. This work was
supported in part by the National Science Foundati'on and by the General
Electric Company.
References
Aryal, M ., Ford, L. H . and Vilenkin , A. ( 1986). Phys. Rev., D34, 2263 .

Arya], M . and Everett, A. E. ( 1 986). Phys. Rev . , D34, 434.
Bach, R. and Weyl, H. ( 1 922). Math. Z., 13, 1 34.
Burden, C. J. ( 1 98 5) . Phys. Lett., 1648, 277.
Davis, R. L . ( 1 98 5) . Phys. Rev., D32, 3 172.
Ford, L. H. and Vilenkin, A . ( 198 1). J. Phys., Al4, 2353.
Garfinkle, D . ( 1985). Phys. Rev., D32 , 986 .
Gott, J . R . ( 1985). Ap. J., 288, 422.
Hiscock , W . A. ( 1985) . Phys. Rev., D31 , 3288.
Hogan, C. J . ( 1 986) . University of Arizona preprint.
Hogan, C. J. and Rees, M. J . ( 1984) . Nature, 3l l , 109.
Israel, W. ( 1 977). Phys. Rev . , 15, 935.
Kaiser, N . and Stebbins, A. ( 1984) . Nature, 310, 1 9 1 .
Kasner, E . ( 1 92 1) . Amer. J . Math . , 43, 2 17 .
Kibble, T. W . B . ( 1976). J. Phys . , A9, 1 387.
Kibble, T. W . B . and Turok , N . ( 1982). Phys. Lett., l l6B, 14 1 .
Linet, B . ( 1 98 5) . Gen. Relativ. Gravit. , 1 7 , 1 109.
Linet, B . ( 1986) . Phys. Rev., 033, 183.

Paczynski, B. ( 1986). Nature, 319, 567 .
Preskill, J. ( 1985). Les H ouches Lectures.
Scherk, J. ( 1975). Rev. Mod. Phys. , 47, 123.
Smith, A . G . ( 1986). Phys. Rev., 0, in press.
Sokolov, D. D. and Starobinsky, A. A. ( 1977). Sov. Phys . Dok/., 22, 3 12.
Turok, N. ( 1983). Phys. Lett., 1 238, 387.
Turok, N. ( 1984). Nuc/. Phys., 8242 , 520.
Vachaspati , T. ( 1986). Nuc/. Phys., 8, in press.
Vachaspati, T. and Vilenkin, A. ( 1985). Phys. Rev., D31 , 3052.
Vilenkin, A. ( 198 1a). Phys. Rev., 023, 852.
Vilenkin , A. ( 198 1b). Phys. Rev. Lett., 46, 1 169, 1496(E).
Vilenkin, A . ( 198 1c). Phys. Lett ., 1078, 47.
Vilenkin, A . ( 1984). Ap. J. Lett., 282, L5 1 .
Vilenkin , A . ( 1985) . Phys. Rep., 121 , 263 .
Vilenkin, A . ( 1 986). Nature, 322, 6 1 3 .
Vilenkin, A . and Vachaspati, T . ( 1986). T o b e published.
Weinberg, S. ( 1972). Gravitation and Cosmology. Wiley : New York.
Witten, E. ( 1984). Phys. Rev., 030, 272.
12
Inflationary cosmology
S T E V E N K . B L A U t and A L A N H . G U T H t
12.1 Introduction
During the last decade, experimental particle physics has tended to confirm
the notion that the standard SU(3) x S U(2) x U ( l ) model accounts for
essentially all the physics that we have seen. In hopes of discovering new
physical laws, many particle theorists have turned to speculations on what
happens beyond the standard model, and much of this speculation has
centered on grand unified theories (GUTs). (For a review, see Langacker,
198 1 .) GUTs explain the quantization of charge and also make a very good
prediction for sin 2 8w ( 8w = Weinberg angle), giving grand unification a
certain amount of plausibility. But the most dramatic predictions of GUTs
occur only at the extraordinary energy scale of 10 1 4 GeV . By the standards
of the local power company, this is not an extraordinary amount of energy
it is roughly what it takes to light a 100 W light bulb for about a minute.
However, the idea of having that much energy on a single elementary
particle is extraordinary. If we were to try to build a 10 1 4 GeV accelerator
with present technology, we could in principle do it , more or less. It would
be a linear accelerator with a length of about one light-year. Now such an
accelerator is unlikely to be funded, so we must tum to other means to see
the 101 4 GeV consequences of GUTs. According to standard cosmology,
the universe had a temperature with k T = 10 1 4 GeV at about 10 - 3 5 s after
t This work is supported in part by funds provided by the Robert A. Welch Foundation
and in part by the US National Science Foundation (NSF) under contract PHY
8304629.
t This w ork is supported in part by funds provided by the US Department of Energy
(DOE) under contract DE-AC02-76ER03069, and in part by the National Aeronautics
and Space Administration (NASA) under grant NAGW-553.
Inflationary cosmology 52S
the big bang, and thus the universe itself becomes the best laboratory for
studying the physics of very high energies.
The interface between particle physics and cosmology has become a very
active field in recent years, and one of the outcomes · has been the
development of inflationary cosmology. The inflationary universe is a
modification of the standard hot big bang model , motivated by several flaws
that emerge when the standard model is extrapolated backward to very
early times. The inflationary model agrees precisely with the standard model
description of the observed universe for all times later than about 10 - 3 o s,
and all the successes of the standa rd model are preserved. For the first
fraction of a second , however, the scenario is dramatically different.
According to the inflationary model , the universe underwent a brief period
of exponential expansion, or inflation, during which its scale factor
increased by a factor perhaps 105 0 times larger than in standard cosmology.
In the course of this spectacular growth spurt all the matter, energy, and
entropy in the universe could have been created from virtually nothing.
The goal of this article is to explain the basics of inflationary models, and
then to summarize some of the more recent research. Emphasis will be
placed on research in which we have been involved - we do not mean to
imply that these investigations are more important than others, but we want
to write about what we understand well. Furthermore, there are already
very good reviews which emphasize other lines of development. Linde's
review ( 1984a) gives particular attention to the recent work in which he has
been involved, such as chaotic inflation, inflation in the context of
supergravity, and also quantum cosmology. There is also an article by Linde
(Chapter 13, this volume) which we have not yet seen . Brandenberger's
review ( 1985) emphasizes the underlying quantum field theory, such as the
effective potential , the finite temperature effective potential , the decay of the
false vacuum, and Hawking radiation in de Sitter space. Steinhardt's article
( 1986) featu res a detailed explanation of the properties that an underlying
particle theory must have in order to be consistent with inflationary
cosmology. Tu rner's 'Inflationary Paradigm ' ( 1985) emphasizes the role of
density fluctuations , the requirements on the underlying particle theory, and
some simple examples of particle theories that work. As the title implies, the
paper also stresses that inflation is not a specific theory, but rather a
mechanism which can be implemented in a number of different ways. We
would also like to recommend the reprint volume edited by Abbott and Pi
( 1986).
Throughout this paper we will set h = c = k = 1, and we will take the GeV
5 26 S. K . Blau and A . H. Guth
as the fundamental unit for everything . We will set G = 1/M� , where MP =

1.22 x 101 9 GeV is the Planck mass. Note that 1 GeV = 1 . 1 6 x 101 3 K =
1.78 x 10 - 2 4 g, that 1 Gev - 1 = 1 . 9 7 x 10 - 1 4 cm = 6.58 x 10 - 2 5 s, and that
1 megaparsec (Mpc) = 3.26 x 10 6 light-yr = 3.09 x 10 2 4 cm = 1.56 x
10 3 8 Gev - 1 . We will use a metric with signature ( - 1 , 1 , 1, 1).
12.2 Summary of the standard cosmological model

There is a rather well-defined standard cosmological model, the basic
features of which will be reviewed in this section. The standard model rests
on the assumption that the universe is homogeneous and isotropic, which
implies that it can be described in comoving coordinates by the Robertson
Walker metric :
ds 2 = - d t 2 + R ' (t { �:,,
1
+ r 2 (d ll2 + sin 2 II d¢ 2) J (2. 1 )
The universe is closed if k > 0, open if k < 0, and flat if k = 0. By convention ,

we rescale r and R(t) so that the parameter k assumes one of the three
discrete values . - 1 , 0, or 1 . We will refer to distances measured by the
coordinate r as 'coordinate' distances, and the corresponding physical
distances will be given by multiplication with the scale factor R(t) . The
expansion of the universe is described by the growth of R(t). The Einstein
field equations relate the behavior of R(t) to the energy-momentum tensor,
which is restricted by the assumptions of homogeneity and isotropy to have
the form Agµv + Buµu v , where uµ = ( l , 0, 0, 0) is a unit vector in the time
direction . The energy-momentum tensor can therefore be written in the
perfect fluid form
(2.2)
In a locally inertial comoving frame this expression reduces to yµv =
diag[p , p, p, p] , so one identifies p as the energy density, p as the pressure,
and uµ as the four-velocity of the fluid . The field equations then imply that
.. 4n
R = 3 G(p + 3p)R ,
- (2 . 3)
(2 .4)
where the dot denotes differentiation with respect to time t. By combining

eqs. (2.3) and (2 .4) , one obtains the relation for conservation of energy,
Inflationary cosmology 527
which can be written as

d d
- (R3 p ) = - p - (R3). (2. 5)
dt dt
The quantity H R/R is known as the Hubble 'constant' even though it
evolves as the universe evolves. Its value today is denoted by
H 0 = h o x ( 100 km/s Mpc)
= h 0 x (9.78 x 109 yr) - 1 , (2.6)
where empirically h 0 probably lies between 0.4 and 1 (Aaronson et al. , 1980 ;
Branch, 1979 ; de Vaucouleurs and Bollinger, 1979 ; Sandage and Tammann,
1976). (We will use the subscript 'O' to denote the present value of any
quantity.)
From eq . (2.4) one can define a 'critical density'
Pc = 3H 2/8nG, (2.7)
which is the energy density that gives a flat (k = 0) universe. Its value today is
given by
(2.8)
The ratio of the actual energy density to the critical mass density (p/Pc) is
called 0, and its value today is believed to lie in the range (Schramm and
Steigman, 198 1 ; Faber and Gallagher, 1979 ; Peebles, 1979)
(2.9)
The universe is assumed to expand adiabatically, with an energy density
which is dominated at early times by- the thermal radiation of effectively
massless (i.e., M � T) particles at temperature T :
n2
p= Ne T 4 , (2. 10)
30 rr
where
(2. 1 1)
Here Nb denotes the number of effectively massless bosonic spin degrees of
freedom (e.g . , photons contribute two units), and Nr denotes the number of
effectively massless fermionic spin degrees of freedom (e.g . electrons and
I
positrons together contribute four units). The entropy density s is given by
(2. 12)
Since the universe expands adiabatically, the entropy per comoving volume
S = R 3s (2. 13)
528 S. K. Blau and A . H. Guth
is conserved. Thus, as long as the number of effectively massless particle
-1.
species does not change, one has
R oc T (2 . 1 4)
Using eq. (2. 10) one finds
p oc R - 4 (radiation-dominated), (2. 1 5)
which can be substituted into the Einstein equation (2.4). For the early
universe R is small enough so that the curvature term - k/R 2 may be
ignored , leading to the simple solution
R oc t 1 1 2 (radiation-dominated). (2. 16)
( ) 1 12 _
It follows that H(t) = 1/2t, and eqs. (2.4) and (2. 10) can then be used to find
45 MP
T2 = (radiation-dominated), (2. 17)
7t Nerr
3 4t
where Mp = 1/G 1 1 2 = 1 .2 x 101 9 GeV is the Planck mass.
At the highest temperatures all of the fundamental particles of nature
contributed to the thermal radiation . As the temperature fell below the mass
of a given particle species, those particles disappeared from the thermal
equilibrium gas - they are sometimes said to have 'frozen out'. Modern
particle physics also predicts that there must have been a number of phase
transitions which took place as the universe cooled, but in the standard
model one assumes that these phase transitions were inconsequential . That
is, they occurred quickly when the temperature fell to the critical
temperature, and the release of latent heat was negligible.
When the temperature reached 1 MeV, the effectively massless degrees of
freedom were the photons, electrons, positrons, and neutrinos. If there are
three species of neutrinos, then Nerr = 1().£ and t = 0. 74 s. Calculations of the
neutrino interaction rates indicate that at this time the neutrinos began to
'decouple' - i .e., they lost thermal contact with the rest of matter. For all
times somewhat later than this, the neutrinos can be described as a
collisionless gas. It follows that the neutrinos maintained a thermal
distribution , with R� equal to the same value it had before decoupling .
When T � ! MeV (at t � 3.0 s), the electron-positron pairs began to freeze
out. Their entropy (given by eq. (2. 1 2) with Nerr = 2) was imparted to the
photons, and as a result R 1'y was increased by a factor of (1J ) 1 13 . The ratio of
the neutrino temperature to the photon temperature,
�/ I'y = (i4f ) 1 / 3
' (2. 18)
has been maintained until the present. If we assume that there are three
species of massless neutrinos, then the radiation energy density after

electron-positron annihilation is given by eq . (2. 10) with
Nerr = 2 + 6 X �( "i) 41 3 3.363.
= (2. 1 9)
1
The interactions which convert protons into neutrons and vice versa all
involve neutrinos, so the ratio of neutrons to protons became essentially
constant as the neutrinos decoupled . After neutrino decoupling, the ratio
was changed only by the free decay of the neutron, which has a mean life of
898 + 16 s. At t � 3 . 5 min the temperature fell low enough for deuterium to
become relatively stable, and the processes of big bang nucleosynthesis
began. It is possible to calculate the resulting abundances of 4He, 3He, 2 H ,
and 7Li, and the comparisons with observation are excellent (Yang et al.,
1984; Boesgaard and Steigman, 1 98 5). The success of these calculations,
along with the observation of the cosmic background radiation, provides
the primary evidence for the standard big bang model of the universe. (Note
that the abundances of heavier elements are not predicted, as these elements
were primarily produced later in the history of the universe in the interior of
stars.)
In addition to thermal radiation, the energy density of the early universe
also contained a contribution from 'matter' - i.e., from massive particles,
such as protons and neutrons. Du ring the time period when these particles
were non-relativistic (i.e. , T <:g, M), their number was conserved and their
contribution to the total energy density was dominated by their rest mass.
This implies that
p oc R - 3 (matter-dominated). (2.20)
For simplicity , we will now assume that the universe is described by Q = i .
If we insert eq. (2.20) into eq . (2.4) and set k = O, we obtain
R oc t 2 1 3 (flat, matter-dominated). (2.2 1)
Today the energy density appears to be almost entirely in the form of
massive particles,t so the density of these massive particles at earlier times
[ ] [
would be given by
J
R(t0) 3 J;, (t) 3
Pm (t) = Pco R(t) � Pco ' (2.22)
J;, o
where t 0 denotes the present time, and J;,0 denotes the present value of the
photon temperature. This expression will equal the energy density P r of
t It is conceivable that the universe today is dominated by some unseen relativistic decay
product, but we will not deal with this possibility here. (See Turner, Steigman and
Krauss, 1984.)
530 S. K. Blau and A. H. Guth
radiation of massless particles when T,, = 5.79 h 57: o 3 eV, where

Yy 0 = 7: 0 x 2.7 K. (2.23)
According to eq. (2. 17) this temperature was reached at time t = 1 .25 x
103 T8h0 4 yr. From that time onward the universe has been matter
dominated .
12.3 Problems of the standard cosmological model

The standard hot big bang model has been accepted by most of the scientific
community, and with good reason. The model is very successful in its
predictions of the Hubble expansion law, the cosmic microwave
background radiation, and the abundances of the light chemical elements.
However, these successes do not probe the behavior of the model all the way
back to time zero. Of the three predictions, the calculation of the element
abundances provides the test of the model which probes it at the earliest time
- the relevant processes began at about one second after the big bang. For
times earlier than one second, there is no direct experimental evidence which
validates the standard model .
When the standard model is extrapolated backward to times much earlier
than one second, there are at least four problems which become apparent . It
is the existence of these problems which motivates the search for a
modification of the model .
The first of these problems is the monopole problem - that is, in the
context of GUTs of particle physics, the standard model leads to a
tremendous overproduction of magnetic monopoles. To see how this comes
about, recall that the monopoles are in fact topologically stable knots in the
Higgs field expectation value.
Suppose the Higgs field is characterized by a correlation length � . Then
for each cubic region of space with edge length � , the Higgs field on one face
of the cube is essentially independent of the Higgs field on the opposite face.
We may estimate that each such cube with volume �3 corresponds to a
topological knot in the Higgs field expectation value, so that the monopole
density is roughly given by (Kibble, 1976)
(3. 1)
While this argument is crude, we nonetheless expect the relation to be
accurate to within one or two orders of magnitude. This accuracy will be
sufficient to establish the conclusion which will follow.
Let � � 101 4 GeV denote the temperature of the GUT phase transition to
the S U(3) x S U(2) x U( l) phase. The corresponding time is obtained from
Inflationary cosmology 53 1
eq . (2. 17) - if we take N err � 10 2 , which is typical of simple GUTs, we find tc �

10- 3 5 s. It can be shown (Ellis and Steigman, 1980) that the Higgs field had
time to reach thermal equilibrium before the phase transition took place, so
that for T > � the correlation length must be given by the only relevant
length scale, r - 1 • (Note that even if one were to imagine that the universe
started out with the Higgs field uniformly aligned , thermal agitation would
destroy this alignment and produce a randomly oriented Higgs field with
correlation length r - 1 . ) When the universe cooled below � ' it suddenly
became thermodynamically favorable for the Higgs field to align uniformly
over large distances . However, it takes time for these correlations to be
established . Causality constrains the Higgs field correlation length to be less
than the horizon distance
IH (t) = R(t)
I' d t
'
,
Jo R(t') (3.2)
which is the total distance a light pulse could have traveled since the initial
singularity. In the standard cosmology R(t) oc. t1 12 , so /H 2t. Since the phase
=
transition is assumed to have occurred quickly at time tc, eq . (3. 1) provides

an approximate lower bound on the magnetic monopole density :
1
n M � � - 3 � -3 . (3.3)
8 tc
Preskill has shown ( 1979; see also Zeldovich and Khlopov, 1978 ; Goldman ,
Kolb and Toussaint, 198 1) that at densities near 1/8t; monopole
antimonopole annihilation is totally ineffectual . Since the monopole mass
mM � 10 1 6 GeV , the monopoles behave as non-relativistic particles with a
number density which is diluted with the expansion of the universe as
1/R 3 (t). The energy density is then given by
P M = n M m M oc. R - 3 . (3.4)
It is useful to evaluate the ratio P M /s ; since the numerator and denominator
each fall off as 1/R3(t) as the universe expands, the ratio remains constant.
The value of the ratio immediately after the phase transition can be bounded
by using eqs. (2. 12), (2. 17), (3 .3), and ( 3 . 4) , which lead to the inequality
m T:
P M /s � c(Nerr) 1 12 M ' ( 3 . 5)
M�
where c 4n 2 (n/45) 1 12 10. 4 .
= =
The present-day entropy density s 0 can be calculated using eq . (2. 12),

assuming that it is dominated by the cosmic background radiation of
photons and neutrinos. Thus

(3 .6)
where N is the number of species of massless neutrinos, each assumed to

v
contribute two spin degrees of freedom, and 4 is the neutrino temperature,

given by eq. (2. 18). Taking N = 3 to reflect the observed families of
v
neutrinos, one finds

s0 = 2.8 x 1 0 3 cm - 3 (3 .7)
and
(3.8)
today.
It may not seem so, but P M is an outrageously high energy density. We
may compare P M to the critical density Pc given by eq . (2 .8), and we find
(taking h 0 � 1)
(3 .9)
which is totally incompatible with the observational limits on the present
value of n, shown in eq . (2.9). In particular, an energy density as high as that
described by eq. (3 .9) would imply that the expansion rate of the universe is
decreasing rapidly - it would have slowed to its present value in only 30 000
years !
The second problem of the standard model is called the flatness problem,
and is based on the realization by Dicke and Peebles ( 1979) that it is rather
puzzling for the value of n today to be of order unity. Q(t) 1 is an unstable
=
equilibrium point in the standard model , and the only time scale which
appears in the equations describing the early universe is the Planck time
t p = (Mp) - 1 = 5.4 x 10 - 44 s. (3 . 10)
Since the universe is about 106 0 Planck times old , it is remarkable that n is
still in the vicinity of one.
Of course for a k = O universe n is always exactly equal to one, but we
regard this possibility as very unlikely. Recall that the Robertson-Walker
metric is defined for any real value of k, and that the discrete choices ( + 1 ,
- 1 , or 0) are obtained by rescaling the coordinates. Thus any positive k is
scaled to + 1 , any negative k is scaled to - 1 , but k = O is obtained only by
starting with k = 0, which is a set of measure zero on the real line. In the
following discussion the possibility k = 0 is ignored.
We can express the flatness problem more quantitatively by considering
the behavior in time of (Q - 1)/Q, which using eqs. (2.4) and (2.7) can be
I nflationary cosmology 533
written as
Q- 1 3k
. � (3 . 1 1 )
8 n Gp R 2
- --
n
The scale factor R can be eliminated from this equation by using (2. 1 3),
leading to
Q- 1 3ks 2 13
- . (3. 12)
n 8nGpS 1 1 3
This equation is useful because the quantities on the right-hand side are easy
to evaluate : p and s are expressed in terms of the temperature by eqs. (2. 10)
and (2. 12), and S is constant as the universe evolves. To calculate the value of
S, note that eqs. (2 .4) and (2.7) imply that
k
R1 = 2 (3 . 1 3)
H (Q - 1) '
and therefore
S= [ �
H ' ( - l) rs (3. 14)
The right-hand side of this equation can be evaluated for the present era,
taking H � ( l0 1 0 yr) - 1 , lf! - 1 1 < 1 , and s = 2.8 x 103 cm - 3 . The result is
s > 108 7 ' (3 . 1 5)
which is an extraordinarily large number. The fact that S is so large is an
alternative statement of the flatness problem. In the context of the standard
cosmological model S is a parameter whose value is fixed by the initial
conditions, and one would expect its value to be of order unity unless there is
some reason to believe otherwise. In this language, the flatness problem is
the fact that the standard model provides no reason to believe otherwise.
One can calculate, for example, the allowed range of n when the
temperature was 1 MeV , when the processes of big bang nucleosynthesis
were just beginning (at about one second after the big bang). Using eq . (3 . 1 2)
one finds
(3. 1 6a)
At the time of the GUT phase transition, when T � 101 4 GeV , one has
1n - l l :S 10 - 49 . (3 . 1 6b)
Finally, if one extrapolates all the way back to the Planck time, when T �
101 9 GeV , then
(3. 16c)
The flatness problem is then the difficulty in understanding why n was so

extraordinarily close to one at these time scales.
Note that the value of n will app roach one at early times in any
Friedmann-Robertson-Walker model which is either radiation or matter
dominated, and so some cosmologists have concluded that there is no
flatness problem. However, if one adopts the point of view that the Planck
time tp is the natural time scale of the system, then t p is not an early time. The
fact that n is so extraordinarily close to one at tp is then not explained by
saying that tp is small, and some other explanation must be sought .
Note also that the flatness problem is not an inconsistency in the standard
cosmological model . If one assumes that the initial value of n was within the
required range, then the model will evolve in a way that agrees with our
observations. The flatness problem is really just a severe shortcoming in the
predictive power of the standard model . The dramatic flatness of the early
universe cannot be predicted by the standard model , but must instead be
assumed in the initial conditions.
The third problem is the horizon problem, first pointed out by Rindler
( 1956; see also Weinberg, 1972, pp. 525-6 ; Misner, Thorne and Wheeler,
1973 , pp . 740 and 8 1 5). The observational basis for this problem is the large
scale homogeneity of the universe, seen most strikingly in the uniformity of
the cosmic background radiation. The effective temperature of the cosmic
background radiation is known to be isotropic to about one part in 103 , and
even this anisotropy can be accounted for by the assumption that the earth is
moving through the background radiation. If one removes a component
which can be attributed to the earth's motion, then the residual anisotropy is
known to be less than one part in 10 4 (Wilkinson, 1986) . This extreme
uniformity is very difficult to understand in the context of the standard
cosmological model , in which the horizon distance (i.e. , the distance that a
light pulse could have traveled since the initial singularity) is rather short .
Consider, for example, two microwave antennae pointing in opposite
directions. Each is receiving radiation that is believed to have been emitted
(or last scattered) at the time of hydrogen recombination, tp about 10 5 years
after the big bang, when the temperature I; was about 4000 K. (At earlier
times the plasma which filled the universe was opaque to this radiation.) At
the time of emission , these two sources were separated from each other by
many horizon distances. To calculate how many, let us assume that n � 1 ,
and that the integral in eq . (3 .2) for the horizon distance i s dominated by the
matter-dominated period, with R(t) = bt 213 for some constant b. The
coordinate distance between the source of the cosmic background radiation
fro bt'dt'2/3 - 3b - 1 Uo1 /3 - t 1 /3

and us is then given by
rcoord - r r ) . (3. 17)
- ft bt' 2/3 - 3b tr .
The coordinate value of the horizon distance at tr is given by
d t' - 1 l/3r
(3. 18)
0
/H.coord
-
Thus, the number of horizon distances separating the two sources m

opposite directions is given by
N=
2rcoord
=2
/H,coord tr
[( ) ] [ ( o ) ]
t o l / 2 - 1 2 7; 1 / 2
= T
- l -:::::, 75. (3 . 19)
The problem is to understand how two regions over 7 5 horizon distances

apart came to be at the same temperature at the same time. t In the standard
model this large-scale homogeneity is simply assumed as an initial
condition.
Like the flatness problem, the horizon problem is not an inconsistency in
the standard model , but represents instead a lack of explanatory power. If
the uni verse is assumed to have begun homogeneously, then it will continue
to evolve homogeneously. The problem is that the very striking large-scale
homogeneity of the universe is not explained or predicted by the model, but
instead must simply be assumed .
The fourth problem is the density fluctuation problem - the difficulty in
understanding the origin of structure in the universe. Although the universe
appears homogeneous on scales greater than a few times 108 light years, on
smaller scales one sees galaxies, clusters, voids, etc. To account for this
structure, one must assume an initial spectrum of mass density
perturbations fJp/p. In the standard model there is no explanation for these
perturbations, and the entire spectrum must be assumed as part of the initial
conditions.
The fact that the spectrum of perturbations is unexplained in the standard
model is a drawback in itself, but the situation becomes even more puzzling
when the starting time is taken as early as t -:::::, 10 - 3 5 s. The problem hinges
on the fact that mass density perturbations are gravitationally unstable - a
t In the above discussion we have assumed for simplicity that n = 1 and that the
universe was matter-dominated throughout the relevant portion of its history. The
calculation has been carried out (Guth , 1983a) without using either of the
approximations, and it was found that the problem is even a little worse - the two
regions were at least 90 horizon distances apart.
536 S. K . Blau and A. H. Gu th
region with excess mass will produce an attractive gravitational field, further
increasing the mass density contrast. Thus, if one extrapolates to very early
times one must find a much more uniform distribution of mass. At t �
10- 3 5 s one must assume that these perturbations were present, but it is
difficult to understand why they were so incredibly small.
To determine how small the perturbations must have been at t 10 - 3 5 s,
=
we must review some of the basic facts concerning the growth of mass
density perturbations (Olson, 1976; Bardeen, 1 980; Press and Vishniac,
1980). Since we are interested in the early universe, it will be sufficient to deal
with perturbations in a flat, radiation-dominated Robertson-Walker
universe. One begins by decomposing the mass density function into
homogeneous and inhomogeneous pieces :
p(x, t) p0(t) + 6p(x, t),
= (3 .20 )
where x is the comoving coordinate of the Robertson-Walker coordinate
system. The next step is to write down the general relativity equations that
describe the evolution of bp and the changes in the metric which are induced.
Since bp/p is assumed to be small in the early universe, one can simplify the
equations by expanding to first order in this quantity. Finally, the linearized
equations can be Fourier transformed, and the resulting equations are then
soluble. The results that one finds depend somewhat on how one chooses to
define the equal-time hypersurfaces in the perturbed spacetime - one
popular choice is the 'comoving gauge', in which the time variable is defined
by a set of clocks that move with the matter fluid . One then finds that the
behavior of the Fourier transformed quantity bp(k, t)/p0 depends on the
relative magnitudes of the physical wavelength and the Hubble length
H - 1 = 2t. (Since the horizon length /H 2 t as well, one often hears the words
=
'Hubble length' and 'horizon length' used interchangeably in this context.)

The physical wavelength corresponding to any given comoving wave
number k grows with the scale factor as Jt, and so at early times the
wavelength is always larger than the Hubble length, and at late times it is
smaller. When the wavelength is much larger than the Hubble length, one
finds that
6p ....
-
p (k, t) oc t, (3 .2 1 )
while the perturbation i s found to oscillate with constant amplitude when

the wavelength becomes much less than the Hubble length. (The
perturbations will start to grow again after the universe becomes matter
dominated, but that era is not relevant for the present discussion .)
Although an elaborate calculation is necessary to verify these results,

there is a simple argument which demonstrates their plausibility. For long
wavelengths the behavior is the same as for an infinite wavelength - i .e., a
homogeneous perturbation from a flat universe. By inserting into eq . (3. 12)
the behavior of s and p for a flat , radiation-dominated universe, one finds
that (Q - 1)/Q grows linearly with t.
Now let us consider in particular the growth of perturbations on the
length scale corresponding to a typical galaxy, with a mass Mg ,...., 101 2 M 0 ,
where M 0 � 2 x 10 33 g is the solar mass. In order for galaxies to have
formed, it is necessary for the perturbations on this scale to have become
non-linear several billion years into the history of the universe, and this
implies that the value of bp/p must have been about 10 - 4 when the
wavelength was equal to the Hubble length (Zeldovich, 1972 ; Press and
Vishniac, 1980). To estimate when this occu rred , one can calculate for any
given time the total baryonic mass in a sphere with radius equal to the
Hubble length. The baryonic mass density varies as 1/R 3 (t), and so it can be
written as
(3 .22)
where H 0, !l80, and R(t0) refer to the present values of the Hubble constant,
the baryonic contribution to n, and the scale factor, respectively. Using the
fact that R T � constant, the last factor can be replaced by T 3 (t)/T 3 (t0), and
T(t) can be evaluated using eq. (2. 17). The baryonic mass within the sphere
is then given by
(3.23)
In the second line above we took the value of Ne rr from eq. (2. 19), which
applies fo r times later than a few seconds. Taking !l80h6 � 10- 2 and setting
M8 = Mg , one finds
(3 .24)
The galaxy scale perturbations will grow linearly with t for times earlier
than t(Mg ), and so they will grow by a factor of about 2 x 1044 between t =
10 - 3 5 s and t(Mg ). Given the desi red value of 1 0 - 4 at t(Mg ), one has
[J - _
_.!!_ (galaxy scale) � 5 x 10 -49 at t = 10- 3 5 s . (3.25)
p
5 38 S. K. Blau and A . H. Guth
This sounds like a very small number, but to decide for sure one should
compare it to some reasonable standard . As a comparison, we will consider
the 1/N 1 ' 2 Poisson fluctuations which are characteristic of any gas of
particles with random locations.
The universe contains about 101 0 particles per baryon - mostly photons
and neutrinos - and so the number of particles associated with a typical
galaxy is given by
N 10 M 0 x
= 1 2
105 7 baryons 10 1 0 particles
M0
x
baryon
.
= 10 7 9 partic1es. (3 .26)
bpp = ---k-
Thus, the Poisson fluctuations in the mass density are
� 3 x 10- 4 0 , (3.27)
N
These Poisson fluctuations are nine orders of magnitude larger than those
permitted by the standard cosmological model. If we extrapolate this model
bp-
back to the Planck time,
_p_ (galaxy) � 5 x 105 7 at t = 10 - 4 3 s, (3 .28)
and so at this time the fluctuations required in the standard model are
seventeen orders of magnitude smaller than Poisson fluctuations.
Thus, if one is to begin the standard cosmological model at times as early
as 10- 3 5 s, then one must begin with density fluctuations which are non
zero, but which are incredibly small - much smaller than 1/N 1 ' 2 . Initial
conditions of this sort seem very peculiar, since density fluctuations of order
1/N 1 ' 2 are almost universal in macroscopic systems - they are present (in flat
space) for a classical gas in thermal equilibrium, and also for quantum
thermal radiation of either bosons or fermions. Density fluctuations are
much less than 1/N 1 12 only for highly ordered systems, such as a
configuration of particles arranged uniformly on a lattice. Density
fluctuations are also much less than 1/N 1 ' 2 for a degenerate Fermi gas at
zero temperature - in this case, the fluctuations are suppressed by the long
range correlations imposed by the Pauli exclusion principle.
The density fluctuation problem may be viewed as a local version of the
flatness problem, and it was described in general terms in the same paper by
Dicke and Peebles ( 1 979). The flatness problem is, in essence, the
observation that the universe in the large is unstable against perturbations
away from n = 1. The density fluctuation problem hinges upon the fact that
the universe is unstable against mass density pertu rbations over a very wide
range of .mass scales.
I njlationary cosmology 53S
12.4 The original inflationary universe

The inflationary universe is a modification of the standard hot big bang
model which avoids the problems discussed above. According to the
inflationary model , the early universe underwent a brief period during which
the matter was in a metastable 'false vacuum' state, driving the evolution of
the universe into exponential expansion. During this period the scale factor
of the universe increased by a tremendous factor - perhaps 105 0 times larger
than in standard cosmology.
There were a number of precursors to the inflationary model , as has
recently been discussed in a very interesting historical article by Lindley
( 1 985). Gliner ( 1965, 1 970) and Gliner and Dymnikova ( 1975) discussed a
state of matter which they called the 'µ-vacuum', which has the same
energy-momentum tensor as the false vacuum. The µ-vacuum, however,
was motivated solely by symmetry considerations. The theory therefore
seemed highly speculative, and a totally ad hoc recipe had to be introduced
to describe the decay of the µ-vacuum . Zeldovich ( 1968) was perhaps the
first to realize that a vacuum energy is equivalent to a cosmological constant
- he suggested that an effective cosmological constant might arise from the
renormalization of the QED vacuum. Linde ( 1974) realized that a
spontaneously broken gauge theory would give rise to an effective
cosmological constant in the symmetric phase, but he concluded (correctly)
that this term could never become dominant at temperatures for which the
symmetric phase is stable. Similar but more detailed derivations were
carried out later by Bludman and Ruderman ( 1977) and by Kolb and
Wolfram ( 1980) . These papers assumed that the universe remained in
thermal equilibrium , and therefore they found no inflation . (Later Guth and
Sher ( 1 983) showed on general grounds that a departure from thermal
equilibrium is a prerequisite for inflation.) Sato ( 198 la, b) became interested
in strongly supercooled phase transitions, attempting to build a consistent
baryon number symmetric universe in which CP-violation occurs
spontaneously in domains of supercluster proportions. At about the same
time Guth and Tye ( 1980) and Einhorn and Sato ( 198 1) studied supercooled
phase transitions as a method of solving the monopole problem. It was this
work that led Guth ( 198 1 ) to the realization that an extremely supercooled
phase transition could potentially solve not only the monopole problem, but
the horizon and flatness problems as well.
In this section we will describe the origi'nal form of the inflationary
universe model (Guth, 198 1). While this model solves the flatness and
540 S. K. Blau and A . H. Gu th
horizon problems, and perhaps the monopole problem as well, it

nonetheless has a fatal flaw, which is sometimes called the 'g raceful exit
problem'. This graceful exit problem is avoided by the 'new inflationary
universe', which will be discussed in Section 12. 7. Here we discuss the
original inflationary universe model for its pedagogical value - the model
demonstrates how the flatness, horizon, and monopole problems can be
solved, and the failure of the model provides the motivation for the special
features of the new inflationary universe.
The original inflationary scenario requires a Higgs field effective potential
V(</>) similar to the one shown in Fig . 12. 1 . While we graph ¢ schematically
as if it were one-dimensional, it is in reality a multicomponent field . At zero
temperature, the global minimum of the effective potential is called the true
vacuum, with ¢ = cPi rue; the scalar field in the universe today is undergoing
small fluctuations about this true vacuum value. The value of V(¢ iru e )
represents the energy density of the vacuum, and is related to the
cosmological constant A by
A = 8nG V( ¢ 1rue). (4 . 1)
The cosmological constant might be large enough to have significant effects
in the present universe, but the mere fact that it does not overwhelm the
present universe implies that the energy density of the vacuum is utterly
negligible compared with the energy densities of relevance to the early
universe. t Thus, we can assume that V( ¢ i rue) = O.
The zero temperature effective potential is also required to have a
secondary local minimum, and the value of ¢ at this minimum will be called
¢raise . The state oflowest energy density for which the Higgs field expectation
value ( ¢ ) = ¢ra ise is called the false vacuum , and the energy density of this
state will be called P r · Furthermore, we assume that there is a critical
temperature 'I'c above which the finite temperature effective potential has a
lower value near the false vacuum than it has near the true vacuum , as
shown in Fig . 12. 1 .t Thus, 'I'c is the critical temperature for a first-order
phase transition .
t While it is clear observationally that the energy density of the vacuum is negligible
compared with that of the early universe, there is no accepted theoretical explanation of
this fact. Naively, one would expect the vacuum energy density to be of order of the
fourth power of the largest mass scale appearing in the theory, which would presumably
mean of order M� . The fact that the empirical limit is more than 120 orders of
magnitude smaller than this theoretical estimate is regarded by many to be one of the
deepest mysteries in physics.
t The finite temperature effective potential is actually the free energy density of the
system, and it must be minimized in the thermal equilibrium state.
For a wide range of parameters (Guth and Tye, 1 9 80 ) the Higgs field in the
minimal S U(5) GUT has an effective potential with precisely these
properties . The false vacuum is a Higgs field configuration which breaks the
gauge symmetry to S U(4) x U( l) rather than S U(3) x S U(2) x U( l ). The
critical temperature is T., � 101 4 GeV .
Although a false vacuum has never been observed, its essential properties
are independent of the details of the particle theory and can therefore be
predicted rather unambiguously. The false vacuum has a constant energy
density p = P r , the value of which is determined by the parameters of the
particle theory. The energy density is typically of the order of the fourth
power of the characteristic mass scale of the theory, which for GUTs means
that
(4.2)
This energy density is almost unimaginably large. It is the energy density
that a large star would have if it were compressed to the size of a proton.
We may deduce the energy-momentum tensor Tµ11 of the Higgs field in the
false vacuum by observing that Tµ 11 is a covariantly conserved tensor
constructed from the Higgs field, the metric, and the first and second
derivatives of these fields. The Higgs field in the false vacuum is constant, so
the only tensors that may be constructed are gµ 11 and
Gµ v = Rµ 11 --J;gµ11 R. (4.3)
Thus, the energy-momentum tensor must assume the form
1'µ 11 = A gµ 11 + BG µ 11 , (4.4)
where A and B are constants. The Einstein field equations are
G µ 11 = - 8nG Tµ11 , (4.5)
Fig . 12 . 1 . A schematic Higgs effective potential appropriate for the original

inflationary universe.
Pc
4>ra1se
542 S. K . Blau and A . H. Guth
so that the term BG µ v in eq . (4.4) may be absorbed into the left-hand side of
(4. 5) by redefining the gravitational constant. Having done this we write
(4.6)
where eq . (2 .2) has been used to identify the constant A with the energy
density of the false vacuum. Eq . ( 4.6) implies that the pressure p of the false
vacuum equals Pr; it is large and negative. Since the pressure is constant, it
-
has no gradient and therefore does not produce any mechanical forces.
However, the pressure does have very dramatic gravitational effects which
will be discussed below. The energy-momentum tensor of eq . (4.6) produces
a term in the Einstein field equations which is identical in form to a positive
cosmological constant - the only difference is that the energy-momentum
tensor of the false vacuum is not permanent, but changes form when the false
vacuum decays.
The false vacuum is unstable, and decays by the spontaneous nucleation
of bubbles of the new phase. For some parameters it is also possible for
magnetic monopoles, which could remain as relics from a high temperature
past, to serve as nucleation sites for bubbles of the new phase (Steinhardt,
198 1a , b; Guth and E. Weinberg, 198 1). However, it appears that a phase
transition drive n by monopoles could never be slow enough to allow for a
successful inflationary model, so we will assume that these monopoles are
stable against nucleation . (Steinhardt ( 198 la, b) has shown that this
assumption is valid for a wide range of parameters in the context of the
minimal S U(5) GUT.) The process of random bubble nucleation will be
discussed in more detail in Section 12.6, but for now it will be sufficient to
simply assume that the nucleation rate is very low.
We are now ready to describe the chronology of the original inflationary
model. As with other cosmological scenarios, the starting point is somewhat
a matter of taste and philosophical prejudice . An advantage of the
inflationary scenario is that it appears to allow a wide variety of starting
configurations - the resulting universe is very insensitive to the details of the
initial conditions. We require only that the early universe was hot (T > I;; ) in
at least some places, and that at least some of these regions were expanding
rapidly enough so that they would cool to I;; before gravitational effects
reversed the expansion . At T,. it is necessary that at least some of these hot
regions had a size about equal to the horizon distance .
If the Higgs field were in thermal equ ilibrium within such a hot region ,
then < </>) � <Praise . In fact, the universe had not had time to thermalize at this
point (Steigman, 1983). Therefore, we need to assume that there were some
Inflationary cosmology 54 �
regions of high energy density with ( </> ) � </> raise . These regions would cool to
� and would then start to supercool below �· At this point the phase
transition would begin, occurring through the spontaneous nucleation of
bubbles of the new phase. We assume, however, that the nucleation rate is
very slow, so that the initially very hot regions would supercool to T near
zero while remaining near the false vacuum. (This assumption is known
(Guth and E. Weinberg, 198 1) to be valid for a wide range of parameters in
the minimal S U(5) model .)
We now follow the evolution of those regions which have supercooled and
approached the false vacuum state. To see what happens next, it is easiest to
begin by assuming that the region is homogeneous, isotropic, and flat.
(Later we will describe what happens when these assumptions are dropped.)
The region can then be described by the metric of eq . (2 . 1 ) with k = 0, and eq .
(2 .4) becomes
(4.7)
which has the solution
R(t) oc ex r (4.8)
(-8nG- Pr) 112 � 10 1 0 GeV � l034 s - 1 .

where
x= (4.9)
3
The remarkable consequence of the huge positive energy density of the false
vacuum is the exponential growth of the scale factor R(t), with minuscule
time constant x - 1 . This exponential expansion is the hallmark of the
inflationary model . The space described by a flat Robertson-Walker metric
with an exponentially growing scale factor is called de Sitter space.t The
dynamics of the inflationary era can also be understood by considering the
second-order Einstein equation , eq . (2.3). This equation implies that the
deceleration of the cosmic scale factor, - R(t), is proportional to p + 3p. In
the standard cosmology p and p are both positive so that gravity acts to slow
down the expansion rate of the universe - R decreases monotonically with
= =
time. During the inflationary era, on the other hand, p -p - pr so that
the pressure contributes a negative term which overwhelms the positive
contribution of the energy density - the effect of gravity is reversed, and the
expansion rate R(t) increases with time. While the inflationary model does
not attempt to explain the formation of the initially hot expanding regions
t The properties of de Sitter space are well described by Hawking and Ellis ( 1973).
544 S. K . Blau and A . H . Guth
which subsequently supercool into the false vacuum state, it does explain the
origin of most of the momentum of the cosmic expansion : the big bang gets
its big push from the false vacuum. The standard cosmology, ·by contrast,
makes no attempt to explain the expansion of the big bang .
Now let us consider what would happen if the initial region were not
homogeneous, isotropic, and flat. In that case, one must examine the
behavior of perturbations about the de Sitter metric. These perturbations
seem to be governed by a 'cosmological no-hair theorem', which states that
whenever the energy-momentum tensor is given by eq. (4.6), any locally
measurable perturbation about the de Sitter metric is damped exponentially
on the time scale of x - 1 • Any initial particle density is diluted to negligibility,
and any initial distortion of the metric is stretched (i .e., redshifted) until it is
no longer locally detectable.
If other matter is present in addition to that described by eq. (4.6), then the
theorem is still applicable provided that the energy-momentum tensor of
the additional matter obeys certain conditions, known as the strong and
dominant energy conditions. The strong energy condition states that
(Tµ 11 -�gµ 11T\) Wµ W" � O (4. 10)
µ
for any timelike vector W , and reduces in the case of a perfect fluid to the
condition p + 3 p � O . The dominant energy condition states that for any
timelike Wµ , Tµ11 Wµ W" � O and Tµ 11 W " is non-spacelike . For a perfect fluid
this condition reduces to p � IPI ·
The no-hair theorem has been demonstrated (Frieman and Will , 1982 ;
Barrow, 1983 ; Boucher and Gibbons, 1983 ; Ginsparg and Perry, 1983) in
the context of linearized perturbation theory, and it is conjectured to hold
even for large perturbations (Gibbons and Hawking, 197 7 ; Hawking and
Moss, 1982). Its validity in the non-perturbative regime has also been
verified in certain exactly soluble models (Wald, 1983). Recently Jensen and
Stein-Schabes ( 1986) have given a proof which holds whenever a
synchronous reference frame exists - i.e. , whenever spacetime can be filled
with a family of timelike geodesics which do not intersect each other. For
non-perturbative situations, the statement of the no-hair conjectu re must
include some kind of restriction to insure that the false vacuum energy
density becomes dominant. In particular, this restriction must exclude the
case of a closed universe that has a positive cosmological constant which is
too weak to prevent it from collapsing. t Although to ou r knowledge this
extra restriction has never been stated, one would guess that it is not
t We thank David Garfinkle for making us aware of this simple example.
stringent. Thus, it seems reasonable to conclude that a smooth de Sitter

metric arises naturally, without any need to fine-tune the initial conditions.t
Since the smoothing effect of de Sitter expansion occurs primarily through
the redshifting of perturbations, it is essential in the arguments for the 'rto
hair' conjecture that the initial perturbations have some minimum
wavelength . In a world described by quantum theory, this assumption about
the initial perturbations is very reasonable. A short wavelength excitation
corresponds to a particle of high energy, so as long as there is some
maximum energy available , there will also be some minimum wavelength.
As the space continues to supercool and exponentially expand , the energy
density is fixed at Pr · Thus, the total energy (i.e., all energy other than
gravitational) is increasing ! If the inflationary model is right , this false
vacuum energy is the source of essentially all the energy in the observed
um verse.
This production of energy seems naively to violate the conservation of
energy, but we must remember that the gravitational field can exchange
energy with the matter fields. The energy-momentum tensor for matter
obeys a covariant conservation equation, which in the Robertson-Walker
metric reduces to eq . (2.5) :
d d
3 p 3
d t (R p ) = - d t ( R ) .
Thus, even in standard cosmology the energy o f matter is not by itself
conserved . Du ring the radiation-dominated period , for example, the
number of particles is conserved as the universe expands, but the energy of
each particle is redshifted as 1/R t . From one second after the big bang until
()
the present, the total matter energy in a comoving volume has fallen by a
factor of about 10 5 . Eq . (2.5) implies that the energy per comoving volume
R 3 p decreases with time when p > O , but it increases with time when p < O.
t Although the authors continue to believe that the 'no-hair' conjecture is valid under
reasonable circumstances, the issues have become controversial . Ford ( 1985) and Barrow
( 1986) have discussed instabilities of de Sitter space, but they deal with situations in
which there is no false vacuum of the type we have discussed . Traschen and Hill ( 1986)
conclude that de Sitter space can be destabilized by minimally coupled scalar fields, but
their method is invalid unless there are N species of such fields, with NH2 � M� . For
typical GUT parameters, this would require an unrealistic 10 1 8 scalar fields. Antoniadis,
lliopoulos and Tomaras ( 1986) have argued that the graviton propagator in de Sitter
space has infrared singularities, but it appears (Allen, 1986) that the pathologies which
they discovered were caused by their choice of gauge conditions. Other papers discussing
instabilities of de Sitter space include Myhrvold ( 1983u, b), Mottola ( 1985, 1986) and
Mazur and Mottola ( 1986).
During the inflationary era one has p = - p = - pr , and one can see that eq.
(2.5) is satisfied identically, with the energy of the expanding gas increasing
due to the negative pressure. If the spacetime were asy mptotically
Minkowskian it would be possible to define a conserved total (i .e. , matter
plus gravitational) energy (see, for example, S. Weinberg, 1972, pp. 165-72;
or Witten , 198 la). The Robertson-Walker metric, however, does not admit
a global conservation law of this type.
Perhaps the most startling idea to come out of recent developments in
cosmology is the suggestion that the universe may possess no conserved
quantities which distinguish it from the vacuum. The total matter energy in
the observed universe is of course very large - � 1078 GeV - but we have just
seen that this quantity is not conserved. The baryon number of the universe
is apparently also huge - � 1078 - but according to GUTs , baryon number is
also not conserved . On the other hand , there are several quantities in nature
which we believe are exactly conserved , such as electric charge and angular
momentum. The consistency of our theories would break down if these
quantities were not conserved . It is therefore very suggestive to note that the
observed value for each of these quantities is compatible with zero . Thus,
provided that baryon ,number is not conserved, the universe appears to be
devoid of all conserved quantities. In that case, it is tempting to believe that
the universe began from nothing, or from almost nothing. The inflationary
universe illustrates the latter possibility. The idea that the universe may have
emerged from 'absolutely nothing' was suggested by Tryon ( 1973) , and
further suggestions along these lines were discussed by Brout , Englert and
Spindel ( 1 979); Gott (1982); Atkatz and Pagels ( 1982); Vilenkin ( 1982,
1983a , b, 1984, 1985b) ; and Linde ( 1983a, 1984b, c).
Let us now return to discuss the fate of those regions which have been
undergoing exponential expansion in the false vacuum state. We suppose
that inflation continued for a time '1.t, during which the regions expanded by
a factor
(4. 1 1)
The inflationary era then ended with a phase transition from the metastable
phase (with ( c/>) � c/>raise ) to the stable phase (with ( c/>) � cl>true ). The original
inflationary universe model relies on the assumption that the phase
transition took place suddenly, with rapid thermalization of the latent heat.
This assumption is now known to be false, and in Section 12.6 we will
discuss what would actually happen. For pedagogical purposes, however,
we will now follow the logic of the original model. We will therefore suppose
that when the energy density P r of the false vacuum was released , it rapidly
thermalized to produce a hot gas of particles - precisely the initial state that
was postulated in the standard cosmological model. Thus, in the
inflationary model the false vacuum energy is also the source of essentially
all the entropy in the observed universe.
The temperature to which this gas reheated can be calculated for any
particular particle theory by using the conservation of energy. Typically one
finds that the reheating temperature is given by
(4. 12)
From here on, the inflationary scenario j oins the standard cosmology. As
the hot gas cooled below the GUT scale, the baryon non-conserving
interactions produced a small net excess of quarks over antiquarks. These
excess quarks eventually resulted in the baryons which we observe in the
universe today (see, for example, Kolb and Tu rner, 1983 ; Yoshimura, 198 1).
Thus, in this model the false vacuum energy is also the source of essentially
all the matter in the universe. Note that the baryon number production must
occur after the inflationary era so that the baryon number density is not
unacceptably diluted .
While the inflationary scenario was certainly motivated by GUTs, it is
worth noting that there are really only two properties of the GUTs that are
essential to allow inflation to take place. First, the theory must contain some
sort of a false vacuum - i.e. , a metastable state with p + 3p < 0 - to drive the
inflation . And , second , the theory must allow for baryon production to take
place after the decay of the false vacuum.
12.5 Successes of the original inflationary model

Let us now discu ss the extent to which the inflationary model provides
solutions to the problems of the standard cosmology that were discussed in
Section 12.3. The monopole problem centers on the fact that the Higgs field
correlation length is bounded from above by the horizon length at the time tr
immediately after the phase transition . In the standard cosmology, the
phase transition is instantaneous, so tr = tc and one has
Jo
ftc dt'
� < l8 (tr) = R ( tc ) R(t') = 2tc . (5. 1)
.
In the inflationary model, on the other hand , the value of R(t) increases by a
factor of Z during the period of inflation , so
"/8 (tr) � 2Ztc . (5.2)
548 S. K . Blau and A . H . Guth
The bound (3 .3) on the number density of magnetic monopoles is then

weakened by a factor of Z3, so the argument which was used to deduce a vast
overproduction of monopoles in the standard model is obviated as l ong as
Z ;<: 104• However, it is not clear in the original inflationary scenario how
many magnetic monopoles would be produced in the phase transition. The
answer to this question depends on the details of the phase transition, which
we will discuss later. Thus, a large discrepancy is replaced by an open
question. The new inflationary scen'.ario provides a much more definitive
solution to the monopole problem, as we will see in Section 12.7.
The flatness problem is solved rather simply by the original inflationary
model . Recall that the problem is solved if we can explain why the
dimensionless parameter S - R3s has such an incredibly large value. To
understand why, note that eq . (4. 12) indicates that J;eheaP the temperature
immediately after the inflationary era, is roughly the same as 'I'c , the value of
the temperature immediately before the inflationary era. The entropy
densities are therefore related by
s(after) ,...., s(before). (5.3)
The scale factors, however, are related by
R(after) � ZR(before), (5 .4)
so that
S(after) ,...., Z3 S(before). (5 .5)
In order to explain why S(after) ;<: 108 7 , assuming that S(before) has a value
of order unity, we require Z ;<: 10 2 9 . This seems like a very large number, but
because the expansion is exponential it is very easy to find parameters (in the
minimal SU(5) model , for example) which give a nucleation rate low enough
to yield this value of Z. In fact, reasonable parameters can lead (Guth and
E. Weinberg, 198 1) to values of Z as high as 10 1 0 10, which is also quite
acceptable ; we require only that S(after) be at least 108 7 - there is no upper
limit. If one does not carefully fine-tune the parameters of the underlying
particle theory, then typically the expansion factor overshoots the minimum
required value, leading to S(after) � 108 7 . Recalling eq . (3 . 12), we see that
this leads to the prediction
n = 1 (today) (5 .6)
to extremely high precision .
The prediction (5 .6) is somewhat problematic, since observations do not
indicate a mass density this large. Observations of the motion of galaxies
within clusters typically indicate a value for n in the range of0. 1--0.3, and big
bang nucleosynthesis calculations (Yang et al. , 1984) place a limit on the
baryonic contribution to n given by

Q B < 0 . 035 h o 2 , (5.7)
where h0 was defined in eq . (2.6). Thus, if inflation is correct then most of the
mass in the universe must somehow remain hidden . Note that it is not
sufficient for the mass merely to be dark - most dark matter candidates that
one might invent would fall into clusters of galaxies and contribute to the
mass that would be measu red by observing the cluster dynamics. The
proposals to account for n 1 include the possibility of dark galaxies
=
associated with 'biased galaxy formation' (Kaiser, 1986 ; Bardeen et al. ,

1986), and the possibility of a smooth component of relativistic particles
created by the decay of some unstable particle {Turner et al. , 1 984; Dicus,
Kolb and Teplitz, 1977). Another possibility is a positive cosmological
constant (Peebles, 1984). Note that the effects of a cosmological constant
can be included m the energy-momentum tensor by setting 8nG Te:c =
- Agµ v, giving
Pvac = A/8nG '
Pvac = - A/8nG. (5.8)
The prediction of inflation can then be written as

n . nma tte r + n vac n ma tter + A/3H 2 1 .
= = (5.9)
A cosmological constant corresponding to n vac � 1 is consistent with
observations, but most particle theorists consider it to be somewhat
implau sible. Naively one would expect that Pvac would be of order M; ,
which would give nvac � 10 1 2 0 . Since some unknown mechanism is
apparently suppressing n vac by 120 orders of magnitude, one would guess
that the mechanism would most likely suppress it by at least a few more
orders of magnitude, making the cosmological constant negligible. This
argument, however, is obviously not compelling.
Retu rning to our discussion of the solutions to the problems of standard
cosmology, we come next to the horizon problem. This problem hinges on
the fact that the standard cosmological model requires that the universe was
homogeneous, at the time of hydrogen recombination, on scales exceeding
75 horizon distances. In the inflationary model , on the other hand, the
horizon distance is stretched during the inflationary era by a factor of Z, and
at all subsequent times the entire observed universe is much smaller than the
horizon distance. Fu rthermore, the inflationary model does more than
simply enlarge the horizon distanc� - it actually provides a mechanism to
create the observed large-scale homogeneity. In the inflationary model the
size of the observable universe at times before the GUT phase transition was
smaller than it would have been in the standard scenario by a factor of Z. if
z > 10 2 5 then the entire observable universe would have been within its
"" '
horizon at t :::::: tc ; it would have become homogeneous at this time by normal

thermal processes, and then this very small homogeneous region would have
been stretched by inflation to become large enough to encompass the
observed universe. The region would then remain homogeneous as it
continued to evolve.
The original inflationary model does not in any way help us to understand
the density perturbation problem . As we will see in the next section, the key
failure of the original inflationary model is the disastrously unrealistic
distribution of mass which it predicts . Fortunately the new inflationary
universe model offers a possible answer to this problem, which we will
discuss in Section 12.7.
12.6 Problems of the original inflationary model

The successes of the original inflationary model depended on the
assumption that the phase transition occurred quickly, with rapid
thermalization of the energy that was released . It is now known that this
assumption is false, and in this section we will discuss what actually happens
when a slow first-order phase transition occurs in an exponentially
expanding space. We will find that the randomness of the bubble nucleation
process leads to gross inhomogeneities, rendering the original inflationary
model untenable. The problems discussed here are avoided by the new
inflationary scenario, which will be discussed in the next section.
The decay of the false vacuum has been studied by Coleman ( 1977) and
Callan and Coleman ( 1977 ; see also Coleman , 1979) , and the description has
been extended to curved spacetime by Coleman and De Luccia ( 1980) . The
false vacuum is classically stable, but it can decay by quantum mechanical
tunneling. The tunneling is a local process, with bubbles of the new phase
nucleating randomly in the midst of the old phase. The o riginal inflationary
model would have worked well if all of the bubbles nucleated at once, but
this is not the case. Rather, the bubbles nucleate with a constant probability
per physical volume per time, which we will call r. (The procedure for
calculating r is discussed in the references cited above.) These bubbles are
approximately spherical ; on the inside < ¢ ) :::::: ¢ t rue , and on the outside < ¢ )
approaches ¢ rais. e as one moves away from the bubble. The nucleation of the
bubble cannot rigorously be described in classical terms, but one can
roughly think of the bubble as materializing at rest . The bubble then grows
at a speed which rapidly approaches the speed of light. Energy is released as

the false vacuum is converted into true, and this energy remains
concentrated in the vicinity of the bubble wall. The energy density in any
region of space inside the bubble will approach zero as the bubble wall
moves outward . If these bubbles were to frequently collide with each other,
then the energy of the bubble walls would thermalize and the original
inflationary cosmology would be workable. As we will see, however, this
does not happen.
To simplify the problem , we will assume that the bubbles materialize with
zero size, and then begin to grow immediately at the speed of light. The
bubbles actually form with an initial size detennined by the microscopic
particle physics ; our approximation will be valid as long as we are interested
in times which are late enough for this initial size to be negligible.
Since the bubble walls expand with the speed of light, the exterior of the
bubble is causally disconnected from the interior. Therefore, the spacetime
outside the bubble continues to be described by the de Sitter metric
d s 2 = - d t 2 + R 2 (t) d x 2 , (6. 1)
with R(t) = e x1 • When viewed from the outside, the bubble walls will travel
along lightlike geodesics in this metric. Thus, if a bubble nucleates at time tN ,
its coordinate radius at a time t > tN is given by
r(t t ) =
' N
I.I R(t')
dt'
--
= x - 1 (e - x1N e x1 ) .
- -
(6.2)
N
As t � oo
(6.3)
The fact that the limit is finite reflects the existence of event horizons in
de Sitter space ; if an event takes place at time t E , a comoving observer whose
coordinate distance from the event is greater than x - 1 e - xiE will never be
able to detect it. The physical horizon distance is then R ( tE ) x - 1 e - XIE = x - 1 '
independent of time. Note that (6.3) does not imply that the bubble stops
growing - it continues to grow at the speed oflight. However, the scale of the
coordinate system is changing so fast that the coordinate velocity of light
approaches zero.
Now consider what happens when random bubble formation begins, at a
time which we will call t8 . The first bubbles which form will have the largest
asymptotic coordinate radii, while the asymptotic coordinate radii of
bubbles which form at a later time t will be smaller by a factor
exp { - x(t - t8 ) } . The asymptotic coordinate volume of bubbles which form
at a later time t will then be smaller by a factor exp{ - 3x(t - t8 )} . However,

the nucleation rate per coordinate volume for the bubbles which form at the
later time is larger by a factor exp{ 3x(t - t8 ) } , since the nucle �tion rate per
physical volume remains constant. Thus there are many more of the smaller
bubbles, and the larger multiplicity compensates exactly for the smaller size.
The total coordinate volume occupied by all bubbles nucleated at time t is
therefore reasonably constant over a very wide range of nucleation times.
The bubbles which form at later times will fill a slightly smaller coordinate
volume because of an exclusion effect - bubbles do not form in regions which
are already in the new phase, so the bubbles which form later have a smaller
region of comoving coordinate space in which to nucleate. This exclusion
effect will be examined quantitatively in the next paragraph. A qualitative
picture of this situation is illustrated in Fig. 12.2.
The bubbles will eventually fill the entire space, in the sense that the
probability that any given point will not be covered by a bubble will
approach zero. To prove this assertion we calculate p(t) , the probability that
an arbitrary point remains outside of all bubbles at time t. Note that p(t) is
the fraction of the comoving coordinate volume available for nucleation at
time t, and th�refore gives a quantitative statement of the exclusion effect
described in the previous paragraph. In calculating p(t) we will use two
simplifications. First, we will ignore the fact that bubbles cannot form inside
of bubbles; instead we will assume that bubbles nucleate at a constant rate
r e3x r per coordinate volume at all points in space . This simplification will
produce no error in p(t) whatever, since any bubble which forms inside of
Fig. 12.2. Random bubble nucleation in an exponentially expanding space.

The largest circles represent those bubbles which nucleate at the earliest
times.
another bubble will remain forever inside it, having no effect on p(t) . Second ,
we will pretend that each bubble appears instantaneously with its
asymptotic coordinate radius x - 1 e - xrN . Since x i is a short time for our
-
purposes , this is a good approximation .

Now consider some arbitrary coordinate volume V and the time interval
between t and t + dt. The expectation value of the number of bubbles that
will nucleate in this spacetime region is given by r v e 3 xr dt. The coordinate
volume of each bubble is (4n/3)x - 3 e - 3 x , so the expected total coordinate
r
volume of the bubbles formed is (4n/3)r vx - 3 dt. Thus, a fraction

(4n/3)rx - 3 d t of the volume is covered with new bubbles. Since the new
bubbles are uncorrelated with any pre-existing bubbles, it follows that given
any region, no matter how chosen , the expectation value for the fraction of
the region that will be covered with bubbles formed during this time interval
is given by (4n/3)rx - 3 d t In particular, it follows that during the time
.
interval between t and t + d t the fraction (4n/3)rx - 3 d t of the region which

was previously uncovered with bubbles will become covered, and this leads
to the exponential form for p(t)
p(t) = exp {- �" < x(t - t 8 ) } (for t > t8 ) , (6.4)
where the dimensionless quantity e is given by

e = r/x4. (6. 5)
Thus, as t � oc , the fraction of space which remains in the metastable phase
goes exponentially to zero. Note, however, that for e < 9/4n the physical
volume of the region that remains in the old phase increases exponentially
with time - the scale factor grows faster than p(t) decreases . For more details
about p(t) , see Guth and Tye ( 1980) and Guth and E. Weinberg ( 198 1 , 1983).
If inflation is to continue long enough to achieve the successes discussed in
the previous section , it is necessary for e to be small. To estimate how small it
must be, we can consider the inflationary era to have ended at a time te when
p(t = te ) reaches some suitably small value Pe � 1 ; the exact choice of Pe is not
important. Then
p(t = t, ) = exp
"
{�
dn Z = p, } (6.6)
where Z = exp { x(te

1029 we require
- te ) } is the inflationary scale factor. In order to have Z ;<:
(6 .7)
The range of values for e which can arise for reasonable values of the
parameters in the minimal S U(5) GUT has been investigated (Guth and
E. Weinberg, 1983). It was found that values anywhere in the range
1 0 - 1 0 000 � 8 � 1 0 1 7 (6.8)
are quite plausible. There are two main reasons why one finds such a
spectacularly large range for 8. It is actually the exponents that appear in eq.
(6.8) which are calculated, and these exponents in turn depend on rather
high powers of the GUT parameters. Thus, values of 8 which satisfy eq . (6.7)
are quite plausible, but it would require significant fine-tuning of parameters
to have 8 near to 10- 2. It is much more plausible to have 8 many orders of
magnitude smaller.
For 8 � 1 , the instantaneous phase transition assumed in the previous
section must be replaced by one in which p(t) approaches zero with an
exponential time constant which is very long compared with the expansion
time x - 1 •
To see if such a slow phase transition might work, one must examine the
properties of the resulting distribution of bubbles (Guth and E. Weinberg,
1983). It can be proven rigorously that if 8 < 10 6 , then the bubbles will form
-
finite sized clusters only, no matter how long one waits, even though
p(t) 0. In mathematical terminology, the bubbles never 'percolate'. It can
�
also be shown that the bubbles do percolate if 8 �0.24. There is some

numerical evidence that the threshold for percolation is at 8 �0.02.
Since the regions that are in the new phase exist as isolated clusters only,
one must look in more detail at these clusters. It can be shown that a typical
cluster is dominated by its largest bubble, with the second largest bubble
being smaller by a factor of order 81'2. For 8 � 1 , this means that a typical
cluster would appear as a single dominant bubble, with an insignificant fuzz
of smaller bubbles around its boundary. The energy of such a cluster would
be concentrated near the wall of the large bubble. Clearly this structure
bears no resemblance to the observed universe.
The problems of the original inflationary universe have also been
analyzed by Hawking, Moss and Stewart (1982). Taking 8 � zh (which is
roughly the maximal value consistent with the inflationary scenario), these
authors calculate that about one-third of all bubbles will grow to a galactic
size (i.e., a size which will evolve to galactic proportions in the present
universe) before undergoing their first collision. They then point out that, in
the standard model, a region of galactic size does not come within the
horizon until· t � 108 s, when T � 106 K. Thus, they argue, the energy in the
wall of such a bubble could not thermalize until T fell below 106 K, a value
too low for baryon production or even nucleosynthesis.
The conclusion is therefore clear that the original inflationary scenario is
unworkable. What is needed is a more graceful way to end the period of
inflation . The model would work if one could find a mechanism which
would allow s to be much less than one during the inflationary era, and then
to simultaneously and quickly change to a large value throughout the space.
No mechanism of this type has ever been found. In the next section we will
discuss the new inflationary model , which provides an elegant solution to
this 'graceful exit' problem.
12. 7 The new inflationary universe

We now turn to the new inflationary universe model , proposed by Linde
( 1 982a, b) and Albrecht and Steinhardt ( 1982). The crucial observation is
that for certain types of scalar field effective potential functions - the
Coleman-Weinberg potential (Coleman and E. Weinberg, 1973) was the
first example to be studied - the phase transition is expected to be quite
different in character from the one which occurs in the original inflationary
model . This new type of phase transition has come to be called a 'slow
rollover' transition , a name which refers to the slow evolution of a scalar
field down a hill in its effective potential diagram. Like the process of
Coleman-Callan tunneling discussed earlier, this type of phase transition is
driven by random processes - in this case it is the random fluctuations of the
scalar field. However, in this model the transition occurs gradually, so that
significant inflation can take place after the phase transition is already well
underway, and after the random choices have been made. The inflation
would then produce huge regions of homogeneous space, and we would be
living today deep within one of these regions. The problems of bubble
coalescence which plagued the original inflationary model do not arise in
this context. The new inflationary model maintains all of the successes of the
original model , and does significantly better. The monopole problem is
clearly solved in the new inflationary model , and the model even offers an
elegant solution to the density fluctuation problem.
We will describe the new inflationary universe model in the context of a
hot initial state. This description follows the original proposal by Linde
( 1982a, b) and Albrecht and Steinhardt (1982), but we want to emphasize
that there are certainly other possibilities. The process of inflation essentially
erases all visible evidence of how it started, so a wide variety of starting
points are possible - the observable predictions of inflation depend only on
how it ends. Some time ago Tryon ( 1973) proposed the speculative but
attractive idea that the universe may have begun from a quantum
fluctuation of 'absolutely nothing'. This idea has since been pu rsued in the
context of the inflationary universe by Vilenkin ( 1982, 1983a, b, 1984, 1985b)
and Linde ( 1983a, 1984b, c). In these scenarios the universe tunnels directly
from a state of 'absolute nothingness' into the false vacuum, with no need for
an intermediate hot phase. In a similar spirit Hartle and Hawking ( 1983)
have proposed a unique wave function for the universe, incorporating
dynamics which leads to an inflationary era. (See also Hawking , 1984a , b,
1986; Moss and Wright, 1984; Hawking and Luttrell, 1984; and Hawking
and Wu , 1985.) Linde ( 1983c, d, 1986a , b, c) has proposed and developed the
idea of chaotic inflation, in which inflation is driven by a scalar field which is
initially chaotic but far from thermal equilibrium. (See also Goncharov and
Linde, 1984a, b, c ; Goncharov, Linde and Vysotskii, 1984; and Khlopov
and Linde, 1984.)
In order for the new inflationary universe scenario to occur, the
underlying particle theory must contain a scalar field cf> which has the
following properties :t
(i) The effective potential function V(cf>) must have a minimum at a value of
cf> not equal to zero .
(ii) V(cf>) must be very flat in the vicinity of cf> = O. The value ¢ = 0 is usually
assumed to be a local maximum of V(cf>).
(iii) At high temperature T, the thermal equilibrium value of cf> (i .e. , the
minimum of the finite temperature effective potential VT(¢ , T)) should
lie at cf> = O.
An effective potential function of this general form is shown in Fig. 12.3. The
point cf> = 0 is an equilibrium point which in the example shown is just barely
unstable. We will refer to the field configuration cf> = 0 at zero temperature as
the false vacuum, even though the term is traditionally reserved for
configurations that are classically stable. The example shown is in fact the
minimal S U(5) Coleman-Weinberg potential (Coleman and E. Weinberg ,
1973), which will be discussed in more detail in Section 12. 10.
The early papers on the new inflationary universe assumed that the Higgs
field responsible for breaking the grand unified symmetry would also play
the role of the field cf> which drives the inflation . However, it was soon found
(as we will discuss in Sections 12.8 and 12. 10) that the effective potential for
t The properties required of the scalar field are described in more detail by Steinhardt
and Turner ( 1984).
the Higgs field is not flat enough, due to the radiative corrections arising
from the gauge couplings. The quantum fluctuations in the Higgs field then
result in mass density fluctuations which are unacceptably large. Thus, in
most newer models the ¢ field is a gauge singlet 'inflation' field, which in
many cases serves no function other than the driving of inflation . Pi ( 1984),
however, has proposed that the axion field can drive inflation, and Ovrut
and Steinhardt ( 1983 , 1984a, b, c, 1986; see also Albrecht and Steinhardt,
198 3 ; and Lindblom, Ovrut and Steinhardt, 1986) have shown that inflation
can be driven by the same scalar field that breaks supersymmetry. We will
use the term 'new inflation' to refer to any model based on a slow rollover
phase transition.
The initial conditions of the universe required by the new inflationary
cosmology are almost identical to those assumed in the original inflationary
scenario. In either case the initial conditions imposed are much less strict
than they are in the standard cosmology. The early universe must have
contained some regions with temperature T > 'I'c, which were expanding
rapidly enough so that they would cool down to 'I'c, before gravitational
1
effects had a chance to reverse the expansion. Such regions must have had a
size of about a horizon length x - when T 'I'c, , or else they would have
=
subsequently collapsed .
Fig. 12.3. The form of the scalar field effective potential function required for
the new inflationary universe model. The example shown is the potential for
the minimal S U(5) theory with Coleman-Weinberg parameters.
10
Pr
'
..,.,-....
>
c.:i
G,)
"'
0
5
.....
"'
:::...
._,
.1-
0 '--��....�� ..
....�--D"'-�-- -'..__�
0 0.5 1 .0 1 . 5
tf> ( l 0 1 5 GeV)
558 S. K . Blau and A. H. Guth
If the scalar field in such a region were in thermal equilibrium , then

(</>) � o. However, the universe would not have had time to thermalize at
this point (Steigman , 1983), so we must assume that there were hot
expanding regions within which < <P) was near to its thermal equilibrium
value of zero. This assumption , however, is not as stringent as it may at first
appear. Thermal equilibrium properties are determined by averaging over
microscopic states, so there are necessarily a large number of microscopic
states with properties similar to thermal equilibrium. In addition, the
Hubble damping (i.e., redshifting) caused by the expansion tends to
suppress perturbations in the scalar field , driving it toward a homogeneous
state with <P � o. Albrecht, Brandenberger and Matzner ( 1985) have carried
out a classical numerical analysis for a variety of initial conditions, and have
found that inflation is quite generic.
As the temperature of a hot expanding region fell below 7;; , the process of
bubble nucleation in principle began . However, the rate for this nucleation
process is typically very low for potential functions with the properties listed
above, and we will assume that it is negligible. The temperature therefore
continued to fall until it reached the order of x, when the situation became
more complicated for at least two reasons. First, curved space corrections to
the quantum field theory became important at this time. Gibbons and
Hawking have shown that quantum fluctuations in de Sitter space mimic
thermal fluctuations at the Gibbons-Hawking temperature
x
Tott = (7. 1 )
2n '
so curved space effects become important when the physical temperature
becomes comparable with this value. In addition, the scalar field was
perched at the top of the plateau in the effective potential diagram .
Depending on the precise choice of parameters, the situation was either
classically unstable or at least very nearly so. Under these circumstances the
approximations used in bubble nucleation calculations break down , and a
more detailed analysis is necessary. These issues will be discussed in Section
12.9, but for now the behavior of the scalar field as it begins to roll down the
hill of the effective potential diagram will be described only in qualitative
terms.
The scalar field hovered for a while in the vicinity of <P 0 , and was then
=
pushed from the top of the plateau by quantum and/or thermal fluctuations.
These fluctuations were random , and therefore the field began to roll down
the hill at different times in different places. The evolution of the scalar field
I njlationary cosmology 559
can be characterized by a coherence length - the distance over which the

field is expected to have been approximately uniform. This coherence length
must be of order x - 1 , since that is the only relevant length scale in the
problem. To see this, note that the scalar field dynamics is governed by the
effective potential function , and that the behavior of small fluctuations
about ¢ = 0 is therefore governed by the effective potential in the vicinity of
¢ = 0. We are assuming, however, that the effective potential is very flat in
this vicinity. M ore specifically, we can assume that at ¢ = 0 one has
av
0¢
0
= , (7.2)
In some models the natural mass scale of the effective potential in the vicinity
of ¢ = 0 might be much less than z, but the Gibbons-Hawking thermal
effects mentioned above would in any case prevent the coherence length
from being greater than a number of order x - 1 . For typical GUT
parameters, x 1 � 1 0 - 2 4 cm
- .
The regions over which the scalar field was approximately uniform will be
called 'coherence regions'. We avoid the use of the word 'bubble' in
describing these regions, for they are non-spherical and they do not have
sharp boundaries.
The initial fluctuations must be described quantum mechanically, but the
subsequent evolution can be described classically. (This transition from a
quantum to a classical description will be discussed in detail in Section 12.9.)
The scalar field within a coherence region was essentially homogeneous, so
we may ignore the gradient terms in the classical equations of motion.
Thermal effects were important in the initial fluctuations, but as inflation
continued the temperature rapidly became negligible. One then has 0¢ =
- o V/oc/J , which in the de Sitter metric of eq . (6. 1) becomes
(7.3)
If one views ¢ as a position coordinate, then this equation describes a

particle rolling in the potential V in the presence of a damping force 3x¢.
The damping term reflects the expansion of the universe. The presence of
this damping term helps to make the new inflationary scenario possible - it
is necessary for the scalar field to roll slowly down the potential so that the
coherence region can undergo a large amount of inflation.
As long as ¢ � o. the energy density p was very nearly equal to Pr · Thus the
exponential expan.s ion continued during this slow-rollover process. The
typical size of a coherence region was then stretched to Zx 1 , where Z = e x 61

-
and 11.. t denotes the time period of the exponential expansion.

When the scalar field </> reached the steep part of the potential , it fell
quickly to the minimum and then oscillated about it. These oscillations were
damped by terms in the equations of motion which arise due to the coupling
of <f> to the other fields in the theory (Albrecht, Steinhardt, Turner and
Wilczek, 1982; Abbott, Farhi and Wise, 1982; Dolgov and Linde, 1982).
Note that the oscillations of </> describe a coherent state of zero momentum
scalar particles - the momentum is zero because the scalar field is
homogeneous. The damping of these oscillations may therefore be
interpreted as the decay of the scalar particles into lighter species. This decay
rate is typically greater than the expansion rate x of the de Sitter space, in
which case there was no significant energy loss to the expansion during the
decay process. The decay products collided with each other and perhaps
decayed into still lighter species, and we will assume that the energy was
rapidly thermalized. The reheating temperature is calculated, just as in the
original inflationary universe, by using conservation of energy. A reheating
temperature of the order of � I'c is again typical.
At this point the scenario joined the standard cosmology, as is the case in
the original inflationary universe. In the standard cosmology the observable
universe had a size of about 10 cm when T was 101 4 GeV, and we therefore
want to be sure that the new inflationary model provides us with regions that
are at least this large. Thus we require zx - 1 � 10 cm, which implies at least
25 orders of magnitude (or 58 exponential time constants) of inflation . Thus
the inflationary era must have lasted at least about 58x - 1 , which for typical
GUT parameters is about 10- 3 2 s. Inflation could, however, have lasted for
much longer than 10 - 3 2 s, in which case the observed universe would be a
relatively small speck within a much larger homogeneous region.
We can now describe how the four problems of the standard cosmological
scenario discussed in Section 12.3 are avoided in the new inflationary
scenario. First, let us consider the monopole problem . Recall that in the
standard scenario, the tremendous excess of monopoles was produced by
the disorder that the Higgs field retained from the previous high
temperature phase (i.e., by the Kibble mechanism). To see how this can be
avoided in the new inflationary model , let us first consider the possibility
that </> and the Higgs field are distinct (as is the case in most recent versions of
the model). In such cases one must arrange for the Higgs field to acquire its
non-zero expectation value either before or during the inflationary era. The
monopole density would then become negligible as it is diluted by the
inflation . Alternatively, we can consider the possibility that inflation was

driven by the Higgs field . (We do not know of any successful models of this
type, but it is not clear that the possibility is completely ruled out .) In such
cases the Higgs field would have been approximately uniform throughout
the coherence region, which was stretched by inflation to become much
larger than the observed universe - it would therefore be unlikely for the
Kibble mechanism to produce even a single monopole in the observed
universe. Some monopoles would still have been produced by thermal
fluctuations after reheating, but this production is suppressed by a large
Boltzmann factor - it would be negligible in the minimal S U(5) model
(Lazarides , Shafi and Trower, 1982 ; Guth, 1983b), and presumably in most
other models as well . It takes much less energy, however, to cause a
fluctuation in the Higgs field as it begins to roll down the hill in the effective
potential diagram. Monopole production by this mechanism may be
significant in the context of the minimal S U(5) model (Moss, 1983, 1984).
However, as we will discuss in Section 12.9, the quantum fluctuations in the
minimal S U(5) model are too large to be consistent with the observed
density fluctuations of the universe. It seems likely that the monopole
production would be totally negligible in any GUT in which the quantum
fluctuations are small enough to be consistent with observation.
The flatness and · horizon problems were solved by the original
inflationary universe, and there is nothing more to be said - the same
mechanisms lead to solutions to these problems in the context of the new
inflationary cosmology. The fourth problem discussed in Section 12.3 was
the density perturbation problem, which will be the topic of the next section .
12.8 Density perturbations in the new inflationary universe

Since inflation stretches microscopic scales into astronomical ones, it
suggests the possibility that the density perturbations which provide the
seeds for galaxy formation might have originated as microscopic quantum
fluctuations. Although it is still not clear if the predictions of the inflationary
model are consistent in detail with the properties of the observed universe ,
we nonetheless feel that this consequence of inflation is very attractive. The
question of density fluctuations was not even considered when the model
was formulated , so it is impressive that the model yields predictions that are
at least reasonable. Furthermore, it must be remembered that the
inflationary model is the first cosmological scenario with enough detail to
even allow a prediction of the density perturbations from first principles.
The question of density perturbations resulting from the new inflationary
scenario was studied in the summer of 1982 by a number of authors at the

Nuffield Workshop on the Very Early Universe in Cambridge, England
(Starobinsky, 1982 ; Guth and Pi, 1982 ; Hawking, 1982 ; Bardeen,
Steinhardt and Turner, 198 3 ; see also Brandenberger, Kahn and Press,
1983). These authors used a variety of methods to treat the problem, but
they all agreed on the answer. Since this time several authors have
questioned the methods used (Hawking and Moss, 198 3 ; Mazenko, Unruh
and Wald, 198 5), but we believe that these questions have been answered
(Brandenberger, 1984 ; Albrecht and Brendenberger, 1985 ; Guth and Pi ,
1985, 1986). The controversies will be discussed further in Section 12 . 10, but
here we will describe the results that were derived at the Nuffield meeting ,
which we believe are correct . In particular we will describe the formalism
used by Guth and Pi ( 1982).
In the new inflationary universe the primordial density perturbations
arise solely from the zero-point fluctuations of the quantized fields.
Although the region which ultimately expanded to become the observed
universe may have contained excitations above the vacuum, these
excitations would not have any significant effect on the present state of the
universe. To see this, note that since the region contained only a finite
amount of energy, quantum theory implies that the excitations must have
had some minimum wavelength. As we discussed in Section 12.4 in the
context of the 'no-hair' conjecture, a sufficiently large amount of inflation
would have redshifted these excitations to immeasurably long wavelengths.
The zero-point fluctuations, on the other hand , have arbitrarily small
wavelengths - indeed they are most significant at very small length scales.
Under ordinary circumstances these zero-point fluctuations are
imperceptible because of their short wavelengths, but the process of inflation
can stretch these wavelengths to macroscopic, and eventually to
astronomical dimensions.
Eq. (7.3) describes the time evolution of a homogeneous scalar field
evolving in the potential V ( ¢ ). We would now like to consider the possibility
that the scalar field might undergo inhomogeneous quantum fluctuations.
The full quantum field ¢(x, t) can be written as
¢ (x, t) = ¢ 0 (t) + 6¢(x , t) , (8 . 1)
where ¢ 0(t) is the classical homogeneous field which obeys eq. (7.3).
Including the·spatial derivative term in 0¢, the equation of motion (using
the de Sitter metric of eq. (6. 1)) becomes
(8 .2)
where a i represents differentiation with respect to the comoving spatial

coordinate xi. Note that the last term denotes the Laplacian with respect to
the physical coordinates . Substituting eq. (8 . 1) into eq . (8 .2) , we obtain
.. . a1 v
<5 ¢ + 3 x&/J = - (¢ ) <5 ¢ + e - 2 x t a f<5 ¢ . (8 .3)
a¢ 2 0
To proceed , we introduce a Fourier expansion with respect to the
comoving coordinates. For an arbitrary functi on f (x, t) we can write
f
f (X , t) = d 3 k e'' · 'j' (k , t). (8 .4)
Eq . (8 .3) then becomes

.. . a1 v
<5 + 3 x<5 = - 2 (¢ o ) <5 {j - e - 2 x tk 2 <5 (j .
($ ¢ (8 .5)
a¢
Note that the second term on the right-hand side tends to stabilize the
fluctuations of the scalar field . It is important at early times, but becomes
unimportant at late times. For the potentials of interest 8 2 V/a¢ 2 < 0 for
¢ � o. so the first term on the right-hand side tends to destabilize scalar field
fluctuations. Thus, the fluctuation amplitude <5 $ (k , t) will oscillate at early
times, and will begin unstable growth at a time t * (k) given by
a1 v
e 2 xt* k 2 = -
-
( ¢o(t * )). (8.6)
a¢ 2
In other words , the physical wavelength at the time the fluctuation begins
unstable growth is given by
2nR(t * ) 2n
A* =
1 _ �
_ �_ (8.7)
k a2 v _ 1 /2 "
__
- ( ¢ o ( t * ))
a¢ 2
From eq. (7.2) it follows that A * � x - 1 .
The history of a fluctuation with a given comoving wave number k is
shown in Fig. 12 .4. The physical wavelength A grows exponentially during
the inflationary era, which extends from ti to tr , and then grows as t 112 du ring
the radiation-dominated era. The graph also shows the Hubble length,
H - 1 (t), which is constant during the inflationary era and grows linearly with
t during the radiation-dominated era. Note that the curves for A and H - 1
cross at two points, labelled A and B the physical wavelength is initially
-
less than the Hubble length, then becomes larger, and then becomes smaller
again . The growth of the fluctuation amplitude becomes unstable at about

the time of point A (the first crossing of the two curves) or perhaps shortly
afterward . Fig. 12.4 is not drawn to scale - for a typical galactic length scale,
the physical wavelength is about 10 2 1 x - 1 at the time of the phase transition;
the time at point B is about 109 s, much longer than the duration of the
inflationary era.
As we stated in Section 12.3 , a density fluctuation bp(k, t)/p in a radiation
dominated flat universe will oscillate with constant amplitude when
A. � H - 1 , after the second crossing of the two curves at B. We will denote this
asymptotic density perturbation a,mplitude by
bp(k)
p H
Our goal will therefore be to calculate the root-mean-square expectation
value for this constant amplitude. We will be interested in wavelengths such
that
x(tr - t * (k)) � 1 (8 .8a)
Fig. 12.4. Wavelengths and the Hubble length as a function of time. One
curve shows the evolution of the physical wavelength A. of a typical
fluctuation, proportional to R(t), and the other curve shows the evolution of
the Hubble length H - 1 (t).
- Inflation -1
I
I
and
(8.8b)
Note that there are no physical processes which operate at scales larger
than the Hubble length, since points at this separation are (at least
temporarily) causally disconnected . Thus the amplitude of the fluctuation
shown in Fig. 12.4 is actually determined by processes that occur in the
vicinity of point A, when the physical wavelength is � x - 1 • The form of Fig.
12.4, therefore, illustrates an important difference between inflationary and
standard cosmology . If Fig . 12.4 had been drawn for standard cosmology,
the physical wavelength would cross the Hubble length only once, and at all
times earlier than the crossing time the physical wavelength would be larger
than the Hubble length. For this reason it is hard even to conceive of a
mechanism , within the standard cosmology, by which the density
perturbations could be determined by physical processes.
Eq . (8.5) is a linear equation of second order. The general solution can
therefore be expressed in terms of two linearly independent basis functions,
and in principle two initial conditions must be given to uniquely specify a
solution . We will be interested in the behavior of the solution, however, only
du ring the period when the scalar field begins to roll off the hill, at t � tr.
Assuming that the time scale for this rolling is large compared with x - 1 , it
will be shown in the Appendix that the damping term in eq . (8 . 5) causes one
of the two linearly independent solutions to become negligible. Specifically,
we will demonstrate that for t > t * (k) one of the two basis functions grows
monotonically, and the other falls off faster than e - 3xr . (To see that this
result is reasonable, the reader may wish to check the simplified case in
which the right-hand side of the equation is replaced by m 2 <5$.) By the
assumption (8.8a) the falling basis function can be neglected, and then the
growing basis function becomes the essentially unique solution. Since the
neglected basis function falls off so rapidly, the expression ' � 1 ' in eq. (8.8a)
can be interpreted roughly as ' ;c; 5'.
For x(t - t * ) � 1 the second term on the right-hand side of eq. (8. 5) can be
neglected , and the equation is then identical to the one satisfied by ¢0(t).
Since we have argued that the solution to the equation is essentially unique,
it follows that <5$(k, t) oc ef>0(t). The proportionality constant may depend on
k, and we give it the name - <5i(k). Thus,
<5 $(k , t) = - <5i(k)</J0(t), (8.9)
from which it follows that
<5¢(x, t) = - br (x)c/>0(t). (8. 10)
To linear order in 6¢, eq . (8 . 1) can then be rewr.itten as

<f> (x , t) = </> 0(t - 6 r (x)) . (8 . 1 1 )
The inhomogeneities in the scalar field <f> (x, t) are therefore characterized by
a single function 6-r(x), which can be interpreted as a time delay function : the
scalar field rolls off the hill at different times in different places. Since the
rolling of the scalar field results in the release of the false vacuum energy
density, this inhomogeneous time delay results in inhomogeneities in the
mass density of the universe.
The completion of the calculation involves two steps : one must calculate
the asymptotic density perturbation amplitude in terms of the time delay
function , and one must estimate the magnitude of the time delay function .
The first step involves a calculation in linearized general relativity. We
will omit the calculation here, referring the reader to the literature which
emerged from the Nuffield Workshop, cited at the beginning of this section.
The result is
6 - (-k )
p = 4x6i(k) . (8 . 12)
H
p
The second _step is the estimate of the magnitude of the time delay function
6i(k ). Since 6i(k ) is a stochastic quantity, we want to estimate something like
its root-mean-square value. There is a small complication, however, arising
from the infinite volume of the space which is used for the calculations. Let
us for the moment consider an arbitrary stochastic function bf (x , t) defined
on a translationally invariant space. We assume that ( bf(x, t)) = 0, and that
the correlations have a finite range :
lim ( 6f(x, t) 6f(O, t)) = O. (8 . 1 3)
lxl -+ x.
One can then see that the expectation value
(bf* (k. , t) bJ(k , t)) = f(�:�, (�:�� r}HX' - Xl ( {!f (X ' , t) bf(X, t)) (8 . 14)
diverges as the volume of space, since the integrand depends only on x ' - x.
This infinity can be factored out by considering the expectation value
(6J* (k ' , t) 6J(k , t)) , which can be written as
(6J* (k' , t) 6J(k , t)) = k - 3 6 3 (k' - k ) [�f(k , t)J 2 , (8 . 1 5)
where
(8 . 1 6)
We therefore adopt the finite quantity D..f(k t) as our root-mean-square

,
measure of fluctuations. The factor k 3 has been inserted in the definition of

D..f (k , t) so that it has the same units at E>f (x, t). The mean-squared
fluctuations at a given point can be expressed in terms of D..f (k) by
integrating with a dimensionless measure of integration :
< [Jf(X , 1)] 2 ) =I�

k
' [aj (k , t )] 2 • (8 . 17)
Note that eq . (8. 13) implies that D..f (k , t) is a local quantity, in the sense that
it depends on the behavior of E>f (x, t) only in the vicinity of a given point ,
taken here to be the origin . We therefore expect that our discussion will be
valid even if the physical region is finite, as long as its size is large compared
with the wavelengths under consideration.
The fluctuations of the scalar field are most easily determined at very early
times , when t � t * (k). (The inequality (8.8b) allows us, however, to restrict
our attention to t � ti, so that the de Sitter metric applies.) At such times the
right-hand side of eq. (8 . 5) is dominated by the second term. If we ignore the
first term on the right-hand side, the equation is precisely that of a massless
(minimally coupled) scalar field in de Sitter space. The two-point correlation
function is therefore given by the propagator of the quantum field theory,
which is well known (Bunch and Davies, 1978 ; Gibbons and Hawking,
1977) :
_ { [
D.. <f> (k , t) = X 2
1+
]} ;
k 2 e - 2 xr 1 2
(t � t * (k)). (8 . 18)
1 6n 3 x2
The derivation of this formula will be sketched in Section 12.9. Notice that
the two terms in the square brackets can be given a simple interpretation.
Since k is the comoving wave number, k e - x r is the physical wave number.
Thus the second term in the square brackets of eq. (8. 18) is precisely the term
that would be calculated in flat space. The first term in the square brackets
represents the Hawking radiation and is most important when k is small,
i.e. , at large wavelengths. One may understand this heuristically by
observing that the large zero-point fluctuations which exist at small
wavelengths are stretched by the exponentially expanding de Sitter space.
Thus at large wavelengths the fluctuations in de Sitter space are significantly
larger than they would be in flat space.
In free field theory approximation , the evolution of the propagator is
determined by the classical equations of motion. (The normalization of the
propagator, however, is proportional to h and is inherently quantum
mechanical.) Thus, we could determine the behavior of D.. </> (k , t) at late times
by following the evolution using eq. (8. 5). Depending on the form of
o 2 V/o<f> 2 (<f>0), this method may or may not be tractable an �lytically - it
would certainly be tractable numerically. One can get a good
approximation to the answer, however, by using a simple matching
argument. While eq. (8 . 18) i's accurate only for t � t * (k), it should be a
reasonable estimate at t � t * (k). Similarly the equation
11 </>(k, t) = 11-r (k) </; 0(t), (8. 19)
which follows from eq. (8 .9), is accurate only for t � t * (k), but it should also
be a reasonable estimate for t � t * (k). Combining this equation with (8. 12),
one finds
11p (k) 4x 11 </>(k, t * (k))
(8 .20)
p H
� cPo (t* (k))
]
Using eqs. (8. 18) and (8 .6) one then has
11p(k) x2 [ 1 a2 v
+ 2 iJ</> 2 o
(<f> (t * (k)))
1 12
(8 .2 1)
p H 1t 3 ! 2 <fi o (t* (k))
� 1
X
In Section 12. 10 we will apply this result to the specific case of the S U(5)
Coleman-Wei_n berg potential.
Since the rollover process must be slow for the new inflationary scenario
to be viable, the function <f>0(t * (k)) must be a slowly varying function of k - it
follows that the fluctuation amplitude given by eq. (8.2 1) is slowly varying. A
spectrum for which [11p (k)/pJ I H = constant is completely scale invariant, and
is known as the Harrison-Zeldovich spectrum (Harrison , 1970; Zeldovich,
1972). Thus, inflationary models yield a nearly Harrison-Zeldovich
spectrum.
Note that the perturbations arise from a nearly free quantum field theory,
and therefore their probability distribution is Gaussian - each Fourier mode
obeys a Gaussian probability distribution, and has a phase and magnitude
uncorrelated with those of the other modes.
The perturbations produced in this way are 'adiabatic', in the sense that
the baryon number to entropy ratio is unperturbed . The adiabaticity i s a
consequence of the short range nature of the baryon production processes,
which occurred in these models very shortly after the GUT phase transition.
The natural length scale for these interactions was necessarily smaller than
the Hubble length, but at this time the perturbations of approximately
galactic scale had wavelengths of order 102 1 times the Hubble length. These
perturbations are described by an inhomogeneous time delay ch(x) of the
phase transition, which occurred in a background geometry which was not
yet significantly perturbed. Thus the time delay function 6-r(x) was
essentially constant within any volume of Hubble size. Any two .distinct
volumes of Hubble size therefore underwent the same evolution, although
slightly out of synchronization. The net baryon number produced per unit of
entropy was determined by this evolution, and is therefore the same
everywhere.
Thus, the generic prediction of inflation is a nearly Harrison-Zeldovich
spectrum of adiabatic, Gaussian fluctuations. A spectrum of this type, with
an amplitude of [L\p(k)/pJ I H � 10 - 4 or 10- 5 , appears to be a viable
candidate for the spectrum that produced the observed structure of the
universe. It is very difficult, however, to know whether this spectrum is
compatible in detail with the observed universe. To answer this question one
would have to calculate the evolution of these perturbations up to present
time - a calculation which requires an understanding of the dark matter in
the universe and the roles of non-linear dynamics and perhaps non
gravitational forces. The difficulties of the calculation can be minimized by
focussing attention on the large-scale structure of the microwave
background radiation (see, for example, Abbott and Wise, 1984a, b), since in
this case a linearized, purely gravitational calculation is expected to be valid.
Unfortunately, though, it is very difficult to obtain data with sufficient
precision. The problem of comparing theory with observation is further
complicated by the need to predict the observable consequences of the
perturbations; i .e., one must understand how enhancements in the matter
density lead eventually to the production of detectable radiation. It has
frequently been assumed naively that 'light follows mass', but now the
possibility of 'biased galaxy formation' (Kaiser, 1986 ; Bardeen et al. , 1986) is
taken quite seriously. The uncertainties about the connection between light
and mass can be bypassed by directly observing the peculiar velocity field
(see, for example, Kaiser, 1983 ; Vittorio and Silk, 198 5 ; Vittorio ,
Juszkiewicz and Davis, 1986; Vittorio and Turner, 1986) , but these
measurements are very difficult. Research on these questions is very active at
present, but it seems too early to expect a definitive answer. There is some
evidence, however, that a Harrison-Zeldovich spectrum of adiabatic,
Gaussian fluctuations cannot account for the detailed structure of the
observed universe. For a sample of the literature, see IAU ( 1986).
The perturbations discussed above are generated by the non-uniformity
of the slow-rollover phase transition, and are expected to be present in any
inflationary model. It is therefore very economical to assume that these
fluctuations are responsible for the formation of structure in the universe. It
is possible, however, that the amplitude [lip(k )/p] IH of these perturbations

was smaller than 10 - 5, and that the perturbations which produced the
structure in the universe arose from another source.
In models that contain axions which dominate the mass of the universe
(Preskill, Wise and Wilczek, 1983; Abbott and Sikivie, 1983 ; Dine and
Fischler, 1983 ; Ipser and Sikivie, 1983), quantum fluctuations in the axion
field would have produced 'isocurvature perturbations', in which the local
axion to photon ratio varied from place to place while the initial mass
density and baryon to photon ratio were uniform (Axenides, Brandenberger
and Turner, 1983; Steinhardt and Turner, 1983; Turner, Wilczek and Zee,
1983; Linde, 1984d, 1985; Seckel and Turner, 1985 ; Turner, 1985, 1986).
Cosmic strings provide another plausible candidate for the origin of
density fluctuations in the universe. (See the review by Vilenkin, 1985a, and
references therein, and also the more recent articles by : Kibble , 1986; Turok
and Brandenberger, 1986; Brandenberger and Turok, 1986; Turok, 1986 ;
Scherrer and Frieman , 1986, Bennett , 1986; Chudnovsky et al. , 1986; Aryal
et al. , 1986 ; Ostriker, Thompson and Witten , 1986.) Strings would also have
an approximately Harrison-Zeldovich spectrum, but the probability
distribution would be distinctly non-Gaussian.
Finally, in the event that the simple Harrison-Zeldovich . spectrum of
adiabatic, Gaussian fluctuations is shown to be unworkable, Silk and
Turner ( 1986) have suggested the possibility of 'double inflation', in which
the amplitude of fluctuations on small length scales is independent of the
amplitude on large scales. In this scenario there is a primary episode of
inflation identical to what we have been discussing, but it is followed by a
second episode of inflation which lasts for only about 38--47 exponential
time constants.
12.9 Quantum theory of the new inflationary universe phase

transition
In the previous section we discussed the quantum fluctuations of the scalar
field c/> which drives inflation , but the analysis rested on one important
simplifying idealization : we assumed that the problem could be described in
terms of local quantum fluctuations about a globally homogeneous classical
solution c/> 0 (t). In this section we will discuss a fully quantum mechanical
treatment of the phase transition in the new inflationary universe, following
the treatment of Guth and Pi ( 1985 , 1986), hereafter called GP. As we will
see, the model of GP is also based on some simplifying idealizations, but we
believe that it qualitatively describes the correct physics and provides
believable answers to the most important questions. We will not present the
calculations here, but we will describe the assumptions and the conclusions.
The idealization of a globally homogeneous classical background
solution is undoubtedly invalid at some level , and a number of questions can
be raised about its use :
(i) Is the picture of a classical slow rollover valid? It has been pointed out
by Mazenko et al. ( 1985) that at high temperatures, when one says that
¢ � o, one really means that the spatial or time average of ¢ is about
equal to zero - the field itself is undergoing large fluctuations. As the
system cools, they argue, it is possible that these fluctuations, which
extend initially out to the minimum of the potential ¢c or beyond , will
cause the scalar field to settle quickly into small regions with ¢ = + ¢c in
each of these regions. When one looks at the spatial average one might
see what appears to be a rolling motion , but the actual local dynamics
could be quite different.
(ii) What is the physical significance of the classical function ¢ 0 ( t) ?
Hawking and M oss ( 1983) point out that the system begins in a thermal
ensemble which possesses an exact symmetry, ¢ � - ¢. The dynamics
is also consistent with this symmetry, and it therefore follows that
< ¢ (x, t)) remains zero for all time - the field presumably does roll down
the hill but, since it is equally likely to roll in any direction, the
expectation value remains zero. A number of authors have suggested
that the quantity { < ¢ 2 (x, t))} 112 can play the role of ¢ 0 ( t) but this
,
identification is problematic. At the quantum level the operator ¢ 2 is

infinite and therefore requires a subtraction which is more or less
arbitrary - it is then not clear how the expectation value of this operator
is related to the behavior of ¢ itself. Furthermore, even for a classical
probability ensemble the quantity { < ¢ 2 (x, t)) } 1 12 does not in general
obey the classical equations of motion for ¢ , so one can hardly expect
the expectation value of a quantum ensemble to have this property.
(iii) How long will ¢ hover in the vicinity of ¢ � o before rolling down the
hill in the effective potential diagram? If one treats ¢ 0 purely classically,
then this question cannot be addressed - there is a classical
homogeneous solution which remains at ¢ = 0 for all time. This
question is crucial, since it determines whether or not sufficient inflation
is obtained. The question has been studied by Linde ( 1982c) and by
Vilenkin and Ford ( 1982 ; see also Vilenkin, 1983c), who calculated the
behavior of ( ¢ 2 (t)) . As we stated above, the required regularization of
this operator makes it somewhat unclear how to interpret the results of
this calculation . The methods discussed below will address the question
in what we feel is a more transparent way. The results obtained with
these newer methods are essentially in agreement with those of the
earlier authors.
In this section we will try to provide answers for these questions. To
understand the quantum mechanical behavior of unstable systems better,
we begin by discussing the relatively simple problem of an upside-down
harmonic oscillator. Consider a single particle of mass m moving in one
dimension , under the influence of the potential
V(x) = -}kx 2 . (9. 1)
We will track the behavior of a wave packet initially having the Gaussian
form
l/J(x, t = 0) ex: exp { - x 2/2h5} . (9.2)
The solution is expressed most simply by introducing the new variables
b 2 = n/(mk) 1 12 (9.3a)
and
W 2 = k/m . (9.3b)
Note that b corresponds to the characteristic quantum mechanical length
scale of the problem, analogous to the Bohr radius of the hydrogen atom.
More precisely, b corresponds to the width of the ground state wave
function of the right-side-up harmonic oscillator potential with the same
absolute value of k. Similarly, w describes the natural frequency of the
corresponding right-side-up harmonic oscillator. The solution maintains
the Gaussian form l/J(x, t) = A(t) exp( - B(t)x 2 ) for all time, where A (t) and
B(t) are complex functions. B(t) can be expressed as
1 .
B(t) = ta n (¢ - 1wt), (9.4)
2b 2
where tan ¢ = b 2 /h 5 . For large times this solution has the form
{
l/J(x, t) ex: exp -
x2 x
2h 2 (t) + 2b 2
i
} (9.5a)
where
(b4 + h o4 ) 1 / 2
h(t) = ewr . (9.Sb)
2h o
The behavior of eq . (9.5b) is easily understood - the width of the wave packet
at large times is minimized for a specific value of h0, which turns out to be b.
If h 0 were chosen very small, then the width would become large due to the
h 0 in the denominator, corresponding to the fact that the uncertainty

principle would require the initial wave packet to have a large spread in
momentum . On the other hand, a large value of h 0 means that the initial
wave packet is already part way down the hill , and therefore spreads
quickly.
The main point in introducing this model is to show that the quantum
mechanical wave function at large times is accurately described by classical
physics. Most textbooks speak of the classical limit of quantum mechanics
in terms of sharply peaked wave packets, which is certainly not the case for
eq . (9.5). However, classical physics also applies whenever the length over
which the phase of the quantum mechanical wave function changes by 2n
(i.e., the de Broglie wavelength) is much smaller than any other length in the
physical problem. In such cases, though, the system cannot be described by
a single classical trajectory. It must instead be described by a classical
probability distribution , which for this system is given by :
f (x, p, t) oc e x 2fh 2 <'> c5(p (mk) 1 ' 2 x) (9.6)
-
- .
The use of the classical probability distribution is justified because the

following two facts can be demonstrated :
(i) f (x, p, t) describes classical physics - i.e., it obeys the classical equations
of motion.
(ii) For any dynamical variable (i .e. , any function Q (x, p)), the expectation
value of Q can be computed by using either the quantum mechanical
wave function I/I or the classical probability distribution f It can be
shown that
( Q) f = (Q )l/t [ l + O(e - 2wt)] . (9.7)
The probability distribution f describes an ensemble of particles, each of
which rolls from rest at x = 0. The various classical trajectories xi(t)
contained in the ensemble differ from each other only by an offset of the time
variable :
(9 . 8)
where the function x ( t) is a solution to the classical equations of motion
which approaches 0 as t -+ oo . Thus, at large times the particle rolls
-
classically down the hill, but the time at which it started to roll is described
by a probability distribution. In this situation an ensemble average (whether
it be an average of x or x 2 ) can obscure the physics, since it averages over
systems that are in very different stages of their evolution.
Having discussed this toy problem, we can now describe the quantum
field theory model proposed by GP. Any perturbative treatment of quantum
field theory (including effective action methods) begins with a free field
theory approximation, but the standard methods were developed for the
purpose of treating perturbations about stable configurations. The GP
model , on the other hand , is a free field theory approximation to a scalar
field in an unstable configuration, perched at the top of a hill in the potential
energy diagram. The model is somewhat crude, but we believe that it
qualitatively describes the correct physics and provides a reasonable
approximation to the behavior of the scalar field in the new inflationary
universe. It is plausible that this model could serve as a valid zero order
approximation to a systematic calculation in which interactions are taken
into account perturbatively - so far, however, no such calculation has been
carried out.
( ) = - !µ2</>2 + -ia</>4 + !!:_4 .

We consider a real scalar field with a potential energy function
2 2
V(<f>) = !a </> 2 _ !!:_ (9.9)
a 4a
The dynamics of the model is simplified by two assumptions. First , we
assume a perfect background de Sitter space, as described by eq. (6. 1). This
is a reasonable approximation for most stages of the rollover. Corrections
due to the transition from the previous hot phase have not been considered.
They are presumably significant, bu t only for the behavior of the very long
wavelength modes. , Modes with periods at the time of the transition which
are short compared with the Hubble time will behave adiabatically and will
will therefore not be excited. Second, we approximate the scalar field as a
free scalar quantum field in de Sitter space, governed by the time-dependent
potential
� 4
Vo (</>) = -1:(µ 2 - !a T 2 )</> 2 + '
a
(9. 10)
where T represents a background temperature obeying the relation

(9 . 1 1)
The value of the parameter f depends on the arbitrary origin of t, and hence
has no physical significance. The temperature-dependent term in eq. (9. 10)
describes the leading high-temperature behavior of the finite temperature
effective potential. To justify its use in this context, GP use an approach due
to S. Weinberg ( 1974) to argue that the validity of perturbation theory
requires that the term be added to the zero-order potential - the negative of
the term is then added to the counterterm part of the Lagrangian , and it is
used in higher orders to cancel high-temperature contributions which would
otherwise cause the pertu rbation expansion to be badly non-convergent. As

we mentioned before, the potential energy function (9. 10) describes a free
field theory which is only an approximation to the truth. In particular, it is
stable for early times and becomes unstable for late times, so </> will roll
indefinitely to larger and larger values. The model will clearly not describe
how </> settles into its minimum, but we believe that it will give a good
description of the slow rollover.
In order to complete the model we must also assume an initial state.
However, in order to describe ou r assumption it will be necessary first to
discuss the behavior of this quantum system for very early times. The
discu ssion will emphasize the fact that, at very short length scales, de Sitter
space is essentially indistinguishable from Minkowski space. For example, it
is conceivable that ou r universe has a positive cosmological constant, and
we might today be living at the beginning of a de Sitter phase with an
exponential time constant of the order of ten billion years - this situation,
however, would in no way affect experiments being done at Fermilab today.
To proceed , we express the scalar field </> as a Fourier expansion :
The mode functions t/J(k, t) then obey the equations of motion :

iii + 3x ifJ [µ 2 - e - 2x' (k 2 + y 2)J t/J,
=
(9. 13 )
where
(9. 14)
As t � - oo , each t/J(k, t) obeys an equation of motion which approaches
that of a simple harmonic oscillator, with angular frequency
w(t) e - x ' (k 2 + y 2 ) 1 ; 2 .
=
(9. 15 )
The frequency of the harmonic oscillator shifts as the universe exponentially
expands, with a rate of order X · However, as t � - oo , the harmonic
oscillator frequency w � oo , and thus for early times w � X· Although the
frequency of the harmonic oscillator i s changing, the change is adiabatic;
each mode then behaves precisely as a mode of a scalar field in Minkowski
space, and we treat it accordingly.
Eq. (9. 1 3) has two linearly independent solutions, each given by the
product of an exponential and a Bessel function. The desired linear
combination can be chosen to match the standard treatment of free field
57 6 S. K. Blau and A . H. Guth
theory in Minkowski space. Hence we choose

t/l (k, t) = -!(n/x ) 1 1 2 e - 3 xr1 2 H�O( z) , (9. 16a)
where H �1> ( z) is the Hankel function (see, for example, Abramowitz and
Stegun, 1972),
P-
- �( ) µ2 1 /2
4 + x2 ' (9. 16b)
and
(9. 16c)
The linear combination in (9. 16) was chosen because at asymptotically early
times it has the form
''' iB(r ) e - iw(t)t
'I' (k ' t)
[ 2 w ( t)] 1 1 2 '
- e e- 3 rx /2 ----,-
(9. 17)
where eiB<r> is a slowly varying phase factor. The factor e - 3 xrt 2 would not be
present in Minkowski space - it is a slowly varying shift in the normalization
conventions. Using the canonical commutation relations with eqs. (9. 16)
and (9. 12), it can be shown that
[d(k), dt(k')] = (P(k' -k). (9. 18)
Given the mode function behavior (9 . 17) and the commutation relations
(9. 18), it follows that at early times the operators dt(k) and d(k) can be
interpreted as creation and annihilation operators of nearly Minkowskian
particles. t
For the special case µ = 0, the Hankel function reduces to the simple
closed form expression
() ( )
H< O1 2 ( z) = - �
1 12
z + i eiz . (9. 19)
3 nz z
Note that the choice of the linear combination of solutions to define the
mode function t/l(k, t) is merely a choice of convention, and does not
constitute a physical assumption. A different choice for t/J(k, t) would lead to
a different meaning for the operators dt(k) and d(k). The statement that the
operators d(k) annihilate the vacuum, on the other hand, is a physical
assumption. For the conventions used here (which are more or less
t To compare the normalization conventions, however, one must be aware of several
differences between this formalism and the standard Minkowski space formalism : in this
formalism t/f(k, t) contains an explicit factor e - 3x11 2 ; the measure o f integration i n (9. 12)
and the J-function in (9. 18) are expressed in terms of coordinate momenta rather than
physical momenta.
I njl.ationary cosmology 577
standard), the state lvac) defined by

d(k ) l vac ) = 0 for all k (9.20)
is the usual Gibbons-Hawking vacuum (Gibbons and Hawking , 1977;
Bunch and Davies, 1978). The propagator for this state is used to derive eq.
(8 . 1 8). Here we are interested , however, in a high-temperature state, so we
assume that each of these harmonic oscillator variables is described at
asymptotically early times by a thermal equilibrium density matrix at
temperature T. Since T is changing with time (eq. (9. 1 1)), one might think
that thermal equilibrium can hold at one time only. However, w is also
changing, and the ratio w/T ---+ constant as t ---+ - oo . This means that
thermal equilibrium can be maintained as t ---+ - oo . Expectation values in
this ensemble can be calculated straightforwardly by using the thermal
ensemble occupation numbers
( d t (k ) d ( k ) ) -
-, - - 1 ,
3 (....k. - -k) (9.2 1)
2 2 1 1
exp{ (k + 1' ) /T } 12 - {J .
For late times it can be shown that the Fourier amplitudes of the field obey a
classical probability distribution, just as the variable x did in the example of
the upside-down harmonic oscillator. In this case the classical
approximation becomes valid for a given mode when its frequency (see eq.
(9. 1 5)) becomes small compared with the de Sitter expansion rate X ·
Since the GP model is a free field theory, essentially everything about it
can be calculated exactly. Here we will summarize the most important
results.
(i) Qualitative behavior of </>(i, t). One might ask what </>(x, t) would
typically look like if one measured it at all points in space at a given time, but
one must remember that a quantum field at a single point is invariably
described by a probability distribution with an infinite width. To obtain a
measurable operator one must 'smear' the quantum field over some finite
volume. One therefore defines a smeared field
$, (X, t) = ( �1 2 1 , fd 3 y exp{ -J(X - Ji) 2/1 2 } cl> (Ji , t) ,

Z n)
(9.22)
where l is a smearing length measured in comoving coordinates. It is then

possible to construct pictures of typical field configu rations by generating a
random value for each Fourier amplitude, using the probability distribution
that has been computed. (To carry this out explicitly one could discretize the
momentum integral by putting the system in a box , with periodic boundary
conditions. ) One can also calculate ( �f(t)) ; this quantity requires no
subtractions , but one must bear in mind the caveal that such an expectation
value averages over systems in different stages of evolution . For ' potentials
A
which are extremely flat (i.e., µ 2 :;;; x 2 and a � 1), the calculation of < </>12 (t)) for
l � 1 / f shows that ¢, (x, t) has a high probability of hovering at ¢ , � o for a
long time before beginning to roll down the hill of V(</J ) . Thus, the picture of
the classical slow rollover is valid under these circumstances, but it would
not be valid for a value of a of order one.
(ii) The meaning of </J0(t). In the new inflationary universe scenario, the
observed universe develops from part of a much larger region which has
evolved into the false vacuum. Let � denote the coordinate diameter of the
region which evolves into the observed universe, and define a 'wave number
of the universe' ku = � - 1 . One can then write a Fourier expansion for the
smeared field, separating it into two parts :
¢, (x, t) = J.
k <ku
d 3k [· · ·J + J.
k>ku
d 3 k [· · -J , (9.23)
where the integrand [ · · "] has the same form as in eq. (9. 12) , but with an
additional damping factor e - < 1 t 2>k 21 2 which arises from the smearing . The
first term on the right-hand side describes fluctuations with wavelengths
longer than the diameter of the observed universe, and thus can be
considered homogeneous for astronomical purposes. It is this term that we
identify with ¢ 0 (t). In this formalism it obeys a classical equation of motion,
but the time at which the rolling begins is described by a probability
distribution. The second term represents inhomogeneities on scales less than
that of the observed universe, and these correspond to the term b ¢(x, t) in
eq . (8 . 1) .
(iii) How long will </J0(t) hover around <P � o? In the GP model this
question can be answered unambiguously by calculating the probability
distribution of ¢0(t). One finds that the probability distribution is Gaussian ,
and that < </J�(t)) � < </Jf(t)) for l = l/f, so the comments made above in
paragraph (i) apply. The GP formalism can also be used to discuss the case
of a Coleman-Weinberg potential . The results are similar to those obtained
by Linde ( 1982c) and Vilenkin and Ford ( 1982), but the picture is somewhat
different. We will return to this discussion in the next section , where we will
discuss the Coleman-Weinberg potential in detail .
(iv) Calculation of density perturbations. The probability distribution for
density perturbations bp/p in the GP model can be calculated exactly. The
results are in excellent agreement with those obtained by applying the
approximate methods described in Section 12.8 to a scalar field with the

potential energy function (9. 10).
12. 10 Inflation in the minimal SU(5) GUT

The new inflationary universe was originally proposed (Linde, 1982a , b;
Albrecht and Steinhardt, 1982 ) in the context of the minimal S U(5) GUT
(Georgi and Glashow , 1974), with the Higgs field playing the role of the
scalar field ¢ which drives the inflation. The Higgs field in this theory
necessarily has two of the three properties which were listed in Section 12 . 7
as requirements for inflation - the minimum of the effective potential
function V(cp) is not equal to zero, and at high temperatures the thermal
equilibrium value of ¢ is equal to zero . The remaining requirement , that
V(cp) be very flat in the vicinity of ¢ = 0, can be obtained by a special choice
of parameters, known as the Coleman-Weinberg values (Coleman and
E. Weinberg , 1973). The effective potential function which was shown in
Fig. 12.3 is in fact the minimal S U(5) Coleman-Weinberg potential .
It is now known , however, that the new inflationary scenario does not
work in its original context - the density fluctuations produced by the
quantum fluctuations of the scalar field have an amplitude which is much
too large. The problem is associated with the fact that although the
Coleman-Weinberg potential is flat in the vicinity of ¢ = 0, it is not flat
enough. When this inconsistency between the new inflationary cosmology
and the minimal S U(5) GUT was first discovered in 1982, it was a serious
disappointment for the people working on inflationary models - at that time
the minimal S U(5) model was considered by many physicists to be by far the
most attractive of the GUTs. Since that time, however, the minimal SU(5)
model has been essentially ruled out by the proton decay experiments, and
there is now no longer any clear favorite among the variety of GUTs that are
consistent with observations. Here we discuss the minimal S U(5) GUT
because it provides a simple context in which to carry out detailed
calculations, and because an understanding of its failure is necessary to
motivate the subsequent developments.
In this section we will make use 6f a number of particle theory results
which we will not derive. We will attempt, however, to state clearly the facts
that we are using, so that the reader can follow our line of reasoning whether
or not he is familiar with the derivations. For the reader who would like
more information about the S U(5) model , we suggest the articles by Georgi
and Glashow ( 1974), Georgi, Quinn and S. Weinberg ( 1974), or Buras, Ellis,
Gaillard and Nanopoulos ( 1978), or the review article by Langacker ( 198 1).
For details about the effective potential function , the reader can refer to the
articles by Coleman and E. Weinberg ( 1973), S. Weinberg ( 1 973), Jackiw
( 1974), and Dolan and Jackiw ( 1974a). For more details about the finite
temperature effective potential, there are articles by Kirzhnits and Linde
(1972), Dolan and Jackiw ( 1974b), S. Weinberg (1974), and Bernard ( 1974).
There is a review article on effective potentials by Coleman ( 1975), and a
review article on finite temperature effects by Linde ( 1979). The review
article by Brandenberger ( 1985) contains a thorough treatment of both the
zero temperature and finite temperature effective potentials.
In the minimal S U(5) theory, the full gauge symmetry is broken to the
subgroup S U(3) x S U(2) x U(l) by a set of Higgs fields <l> which transform
according to the adjoint representation of S U(5). That is, <l> represents a
traceless, hermition, 5 x 5 matrix of fields (with 24 independent components)
which transforms under the gauge group as
<l>'(x) = g - 1 (x)<l>(x)g(x) , ( 10. 1)
where g(x) denotes an S U(5) matrix. The symmetry breaking is
accomplished by the fields <l> acquiring a vacuum expectation value of the
form
<l> - (__l_ /
1 ) l 2 � d"iag [ 1 , 1 ' 1 ' _J_2 ' - _J_J ( 10.2)
- s If' 2 ,
_
or a form which is gauge-equivalent to this. The prefactor (/5 )112 is chosen so

that
( 10.3)
The space of generators of the unbroken subgroup are the hermitian
traceless matrices which commute with the matrix displayed in ( 10.2) : the
matrices with non-zero entries in the upper left 3 x 3 block generate an S U(3)
subgroup, those with non-zero entries in the lower right 2 x 2 block generate
an S U(2) subgroup, and the matrix in ( 10.2) generates a U(l) subgroup .
By restricting <l> to the form of eq. ( 10.2), the effective potential becomes a
function of ¢ . The Coleman-Weinberg condition is then
a2 v
aIf'� 2 <f> = O = 0. ( 10.4)
Computing the effective potential including the one-loop radiative
corrections for the gauge interactions, one finds at zero temperature
V( </J) = i�a 2 [ ¢4 ln( ¢ 2/a 2) + ! ( a 4 - ¢4 )] , ( 10.5)
where a = g2/4n � J5 , where g is the gauge field coupling constant, and a �
1 .2 x 10 1 5 GeV. The form of this potential was shown in Fig . 12.3. The
minimum lies at ¢ = a, corresponding to the true vacuum. The point ¢ = 0 is

an equilibrium point which is just barely unstable. The configuration with
¢ = 0 at zero temperature will be called the false vacuum .
The finite temperatu re effective potential is given by
V, (¢ , T) = V(¢) + 1!?' r ( 5"

q2 dq In 1 - exp - q 2 + rx ¢ 2
3
1'2
{( ) / T}}
( 10 .6)
The behaviour of V (cp, T) is shown in Fig . 12.5. The curves show a first
order phase transition at T = T.: = 1 .25 x 10 1 4 GeV . For T < T.: the minimum
of the effective potential is in the vicinity of ¢ = a, while for T > � the
minimum lies at ¢ = 0. For all temperatures 0 < T < � there is a metastable
phase with ¢ = 0, corresponding to a local minimum of the effective
Fig. 12.5. The Coleman-Weinberg finite temperature effective potential for

the minimal S U(S) Higgs field .
0
-
>
<l"
C-'
II)
0
'°
"'
::::..
.._..
-5
1 .5 X 1 0 1 4 GeV
-10
· o.s 1 .0 1 .5
5
cf> ( l 0 1 GeV)
potential . For small T the dip which creates the minimum at ¢ 0 has a
depth of order T 4 , and a width of order T.
=
Having summarized the particle physics background, we are now pre
pared to discuss the details of the inflationary scenario in the context of the
minimal S U(5) GUT. Following the general scenario laid out in Section
12.7, we assume that the early universe contained hot expanding regions
within which ( ¢) was near to its thermal equilibrium value of zero. As the
temperature of such a region fell below 1'c, it became possible for bubbles of
the new phase to nucleate. However, as long as T � x � 10 1 0 GeV one can
rely on flat space calculations (Sher, 198 1 ; see also Abbott, 198 1 ; Billoire
and Tamvakis, 1982 ; and Cook and Mahanthappa, 1982) that show that the
nucleation rate is completely negligible.
Once the temperature fell to the order of x, the scalar field was agitated by
quantum and thermal fluctuations which tended to start it rolling down the
hill of the potential energy diagram. It is important to know whether the
scalar field would remain on the top of the hill for long enough to allow for
sufficient inflation. This question has been addressed by Linde ( 1982c) and
Vilenkin and Ford (1982), who concluded that this model does not lead to
enough inflation . It will be easier to discuss the issues, however, after more
groundwork has been laid, so we will return to this topic at the end of the
section. Meanwhile, in order to continue our pedagogical discussion, we will
assume that the scalar field remained perched at the top of the hill of the
potential energy diagram for some unspecified length of time.
The scalar field then rolled down the hill of the effective potential diagram,
obeying the classical equation of motion. The general form of this equation
was given as (7.3), which for the case of the Coleman-Weinberg potential
becomes
( 10 . 7)
The logarithm is a slowly varying function compared with ¢ 3 . The coupling
constant ex may also be viewed as a function of ¢ , because it is a 'running'
coupling constant (Sher, 198 1) whose value depends on the choice of a
renormalization scale µ :
ex (µ 2) = �1
4n
3 n (µ ; S U(5) )
2 A2 '
( 10.8)
=
where A s u ( S) 2.5 x 105 GeV is fixed by the requirement that ex � ls at the
GUT scale. t The renormalization scale is in principle arbitrary, but the
t We have used here the renormalization group equations for the unbroken S U(5) gauge
group , since we are interested in the case </> �o.
choice µ = ¢ is preferred - it helps to keep the contributions of higher-order

perturbation theory small by avoiding the occurrence of large logarithmic
factors. This dependence on ¢ , however, is also slowly varying compared
with ¢ 3 . Thus we may approximate a 2 j ln ( ¢ 2/u 2)j as a constant and write
- a </Jv = a <P 3 , ( 10.9)
8
where a � 1- for ¢ � x, the only scale relevant to the problem. As we will see, it
is also safe to ignore the (/j term in eq . ( 10.7), which is then solved by
2 3x
<P o (t) = 2a(tr - t) (10. 10)
We have chosen the constant of integration so that ¢ 0 --+ oo at a time which
we call tr. Of course the approximation ( 10.9) breaks down when ¢ 0
becomes comparable with u, but tr can still be thought of as the time when ¢ 0
becomes large. The neglect of the (/j term can now be justified self
consistently, provided that x(tr - t) � 1 . Using ( 10. 10) one has
¢0 1
- � 1. ( 10. 1 1)
3x <P o 2x(tr - t)
A plot of eq . ( 10. 10) is shown in Fig . 12 .6. The field ¢ 0 changes almost
imperceptibly until x(tr - t) � 10 or so, and then approaches infinity very
sharply. If the ¢-axis had been drawn with a linear scale, the motion of ¢ 0
would have been imperceptible until x(tr - t) � l , at which time the
approximations become invalid.
As long as ¢ 0 � u , the energy density p was very nearly equal to P r · Thus
Fig . 12.6. The evolution of the scalar field during a slow-rollover phase
transition in the minimal S U(5) theory.
,.-..
> 1 Q13
C-'
d)
'-'
,.-..
.....
'-'
� 1 011
x Ctr -t)
the exponential expansion continued during this slow-rollover process. For

the scenario to work , it is necessary for the length scale of homogeneity to
have stretched from x - 1 to at least about 10 cm before the scalar field <P o
rolled off the plateau in Fig . 12.3. This corresponds to an expansion factor of
about 10 2 5 , which requires about In 10 2 5 = 58 time constants (x - 1 ) of
expansion . Thus, the scenario will work if the classical rolling described by
eq. (10. 10) can be considered a valid approximation for times as early as
x(tr - t) � 58 , which according to (10. 10) implies a value of ¢0 as small as
0.23x. On dimensional grounds one expects the initial fluctuations to have
been of order x, so detailed calculations are necessary to decide if they were
large or small compared with 0.23x.
We are now prepared to use the formalism of Section 12.8 to determine
the spectrum of primordial density fluctuations [L\p (k)/p] IH . The quantity
t * (k) is given by eq. (8.6), which can be evaluated for the Coleman-Weinberg
potential by using eqs. ( 10.9) and ( 10. 10). One then has
1 xtr 2
x(tr r*) 9(xk - e )
e2 - ( 10. 12)
2x(tr - t * )
In order to use this relation we will need to express the comoving
coordinate wave number k in terms of the physical distance scale measured
at the present time, 10. At the time tr of the phase transition the universe was
reheated to some temperature 'I;eheat � 10 1 4 GeV , so taking 1R � constant
and using T0 � 3 K � 3 x 10 - 1 3 GeV to denote the current temperature, one
has
lo = R(t o )k - 1 =
R (tr)
k - e "' (
R(to ) l xrr - 7;eheat 1
To X
)
- (Xk - 1 ex tr ) . (10. 1 3)
Thus
(10. 14)
where
b = r�:at X - l � lO m � 10 - 1 5 Jigh t-yr.
T.
( 10. 1 5)
A galactic scale corresponds to about 10 6 light-yr, so 10/b � 10 2 1 .
Eq. (10. 12) can now be rewritten as
x(tr - t * ) = ln ( l 0/b) - 1 I n[�x(tr - t * )] . (10. 16)
For wave numbers corresponding to the scale of galaxies the second term on
the right-hand side is a factor of 40 smaller than the first term, and it will
therefore be neglected in what follows. The fluctuation amplitude as
expressed by eq. (8 .2 1) can then be evaluated using eqs. ( 10. 10), ( 10. 16), and
( 10.9), with the resultt

:
!ip k)
H
�
u:, r z In 3 / 2(/ ofbl
In deriving ( 10. 17) the second term inside the square brackets on the right
( 10. 1 7)
hand side of eq. (8 .2 1) was neglected - on the scale of galaxies this term has a
value of 0.09, which is small compared with one.
Inserting numbers into ( 10. 17), one finds that on the scale of galaxies
p
/). � 70 (/ � 106 1ight-yr).
- 0 ( 10. 18)
p H
The density fluctuation amplitude varies slowly with wavelength - for the
scale of the observed universe one finds
p/). � 9 1
- (/ 0 � 10 1 0 light-yr) . ( 10. 19)
p H
These values of (!1p/p )IH are much too large to be acceptable - the desired
number would be about 10- 4 . In fact, the numbers obtained here are so
large that the predictions of the theory are uncertain. The linearized
calculation which we have carried out requires that (!1p/p)IH � 1 , so the
calculation is not valid. The calculation suggests, however, that the density
perturbations are strongly non-linear at the time when the wavelength
becomes comparable with the Hubble length , so it seems reasonable to
speculate that the matter would collapse to black holes on approximately
this scale. Note that in a flat universe any sphere with a radius r equal to the
Hubble length encloses a mass M = -j:nr3 p such that r = rschwarzschild = 2GM. In
an unperturbed flat universe the momentum of the expansion prevents the
collapse of such a sphere to a black hole, but one would expect that any
increase in mass density by a factor of 2 or more would result in black hole
•
formation. Since (!1p/p)IH � 1 on all relevant scales, we expect that the black
holes would continually coalesce to produce larger black holes on the scale
of the ever-increasing Hubble length.
In any case it seems clear that the theory is inconsistent with the observed
universe. If the true prediction of the theory were (!1p/p )IH � io - 4 , then the
linearized calculation would have been valid and could not have led to the
results ( 10. 18) and ( 10. 19).
Note that the desired answer of about io- 4 could be obtained if a were
t This formula is a factor of J2 larger than that in Guth and Pi ( 1 982) because here we
derive the asymptotic amplitude of the wave, while the earlier reference derived the
asymptotic root-mean-square value.
586 S. K . Bla u and A . H. Guth
smaller, but it would have to be very much smaller - the required value is
about 10 - 1 2. Thus, it seems that an extremely weakly interacting scalar field
is needed to drive inflation.
We can now return to the question of whether or not the Higgs field in this
model would remain perched at the top of the hill for long enough to provide
an adequate amount of inflation. As mentioned above, this question has
been studied by Linde ( 1982c) and Vilenkin and Ford ( 1982 ; see also
Vilenkin, 1983c), who found a negative answer. Their argument begins with
the observation that for ¢ � o , the Coleman-Weinberg potential can be
approximated by V( ¢) = constant, describing a free , massless scalar field in
de Sitter space (minimally coupled to gravity). This quantum field theory
has the peculiar property that there exists no stable vacuum state - the scalar
field never settles near any particular value, but instead undergoes an
endless random walk up and down the real number line. Since there is no
stable vacuum the renormalized expectation value < ¢ 2 ) is time-dependent,
and is found to behave as
1 3
(¢ 2 (t)) = X (t - t o ) , ( 10.20)
4rc2
where t0 depends on the arbitrary infinite subtraction. If one interprets
< ¢ 2 (t)) as the quantum version of the classical ¢6(t) (an interpretation
which we do not recommend), then one can compare the time derivative of
( 10.20) with the time derivative of ( 10. 10). If one insists that d ( ¢ 2 (t))/d t ;$
d¢6/dt, one finds that
( )
6 rc2 1 12
x(tr - t) ;$ -----;;- � 1 1. ( 10.2 1)
Thus, the growth in ¢2 due to quantum fluctuations ( 10.20) will be smaller

than the growth due to classical motion ( 10. 10) only during the last 1 1
exponential time constants of the slow rollover. Since adequate inflation
requires the classical solution to be valid for at least 58 time constants, it
seems clear that the scalar field does not remain long enough near the top of
the hill .
The formalism of GP described in Section 12.9 leads to a conclusion
similar to the pioneering results of Linde, Vilenkin , and Ford, but the
picture appears slightly different. To approximate the minimal S U(5)
Coleman-Weinberg situation within the GP formalism, we take µ 2 = 0 - this
gives the zero temperature potential V(</J) = constant. To determine a , we
examine the high-temperature behavior (Dolan and Jackiw , 1974b) of the
minimal SU(5) finite temperature effective potential (eq. ( 10.6)) :

Vy (¢ , T) � V(¢) -
2n 2 4 Sn
S rx
T + 2 ¢2 T2 + . . " ( 10.22)
Comparing the coefficient of T 2 with eqs. (9. 10) and (9. 14), one sets
a = 20mx � l .4
y2 = ia f 2 �o.35f 2 .
.
( 10 2 3a )
( 10.23b)
(Note that the term in eq. ( 10.22) proportional to T 4 is independent of ¢ ,
and therefore has n o dynamical effect . The term i s not included i n the GP
calculation.)
As explained in the previous section, GP define the smeared field ¢,(x, t),
and then calculate the finite quantity ( $,2 (x, t)) . This quantity averages over
systems in different stages of evolution, and therefore the interpretation has
to be done carefully - nonetheless, a small value for the quantity implies
unambiguously that quantum fluctuations are small . The calculation is
carried out using eqs. (9. 12), (9.22), (9 . 16), and (9.2 1), and then eq. (9. 19) is
oo -k2[2 {
< <P f (x t)) = 4n 2 o k 2 + y2 3 / 2 1
A
+
k2 + y2 }
used to reduce the answer to the special case µ = 0. The result is given by
, x2 f k2 dk e
e - 2xr
J
x coth
{( ) }
( )
k 2 + y 2 1 /2
.
x 2
( 10.24)
2T
As an example, we can take the co moving smearing length as l = 1/f, so the
physical smearing length at any 'time is 1/T, where T = f e - xr is the
background temperature. On dimensional grounds, one expects this
smearing length to correspond roughly to a coherence length. The integral
can then be carried out numerically, yielding
< ¢f(t)) = 0.02 19 x 2 + 0.0 1 9s r 2 • ( 10.25)
Thus, the mean-squared value of the smeared field consists of a thermal
contribution which redshifts to zero, and a residual contribution. Note that
( (fif(t)) does not contain a contribution that grows linearly with t, as in eq.
-
( 10.20). The difference is in the nature of the short distance cutoff. Eq. ( 10.24)
is expressed in terms of a fixed coordinate smearing length l this type of
smearing is appropriate to discuss the behavior of the mean value of ¢
throughout the region which will evolve to become the observed universe.
The renormalized expectation value ( ¢ 2 ) , on the other hand, is calculated
with a fixed physical cutoff momentum - this type of cutoff is appropriate to
discuss local measurements of ¢, where the short wavelength fluctuations
are operationally regulated by the resolution of the measuring device. If eq .

(10.24) is evaluated for a fixed physical smearing length, so that. I = /phys e - xt,
then one can show that at large times the right-hand side grows linearly with
time, with the same coefficient as in eq . ( 10.20).
Eq. ( 10.25) shows that $,(t) typically remains rather small. Note that the
residual contribution corresponds to a root-mean-square value of
j(0.02 19)x = 0. 1 5x, which is smaller than the value of 0.23 x at which the
classical equations of motion are required to be valid.
Similarly we can study the quantity ¢ 0 (t), as described at the end of
=
Section 12.9. Taking � 1/f as the coordinate diameter of the region
hypothesized to evolve into the observed universe, and taking I � � we can
calculate ( ¢6(t) ) . The result is an integral of the same form as ( 10.24), except
'
that I � o and the upper limit of integration is f. Again the integration can be
carried out numerically, with the result
< ¢6{t) ) = 0.0212x2 +o.02 10r2. ( 10.26)
Here the residual root-mean-square value is 0. 16x - again acceptably small.
Thus, quantum and thermal fluctuations do not cause the mean value of <P
on large scales to fluctuate to a value large enough to upset the validity of the
classical equations of motion . However, the fluctuations 6 ¢(x, t) on smaller
scales are very large, completely destroying the validity of the slow-rollover
picture. The problem can be seen by calculating ( [b¢(x, t)] 2 ) , using eqs.
(8. 1 7) and (8. 18). If we assume that the second term in the square brackets of
eq. (8 . 1 8) is negligible (as is the case when the physical wavelength is large
compared with the Hubble length), and if we assume that the integral in
(8. 17) receives contributions from wave numbers in a finite range km in < k <
kmax , then we find
( [6 ¢(.X , t)J 2 > = x

4n 2
( )
2 1n kmax
km in
. ( 10.27)
Using the decomposition ¢ = ¢0 + b¢ of eq .
= 1.
(9.23), one has a time
independent lower limit of integration, km in ku = � - The integral of eq .
(8 . 1 7) should be cut off at large k by the smearing function, which gives a
factor e - k2 1 2 (as in eq. ( 10.24)). For present purposes, however, we will
approximate the smearing by a sharp cutoff at some physical wave number
kphys, max . Then kmax = ex ' kphys, max , and the right-hand side of eq. ( 10.27) is seen
to have the same behavior as the right-hand side of eq. ( 10.20). The
significance of the numerical result obtained in eq . ( 10.2 1) can now be
restated : the growth of the classical solution ¢ 0 (t) will exceed the growth in
the root-mean-square value of b</J(x, t) only during the last 1 1 exponential
time constants of the slow rollover. The assumption that J¢(x, t) is a small
pertu rbation that can be treated linearly is therefore badly violated , and the
entire picture of the slow rollover breaks down . Thus, this model seems to
exhibit the pathological behavior suggested by Mazenko, Unruh and Wald
( 1985).
We conclude that the new inflationary model with a minimal S U(5) GUT
has two flaws : the density fluctuations are unacceptably large, and the
amount of inflation is insufficient. Both of these flaws, however, are
symptoms of one underlying problem : the quantum fluctuations in J¢(x, t)
are too large. As we saw in eq . ( 10.27), however, the size of these quantum
fluctuations is controlled by general principles of quantum field theory, and
does not depend on the detailed shape of the minimal S U(5) Coleman
Weinberg potential. What is needed, then , is a potential for which the
classical solution gives a value of cfi(t) large enough to suppress the effect of
these fluctuations. The example studied here suggests that the requirement
of obtaining [�p(k)/pJ IH � 10 - 4 is more stringent than the requirement of
obtaining su fficient inflation. Thus, it seems plausible that in models for
which [�p(k)/pJ IH has an acceptable value, the expected amount of inflation
will greatly exceed the minimal requirement of 58 time constants.
Once the failure of the minimal S U(5) theory was discovered, efforts
turned toward the construction of particle theories that have the properties
necessary for the new inflationary cosmology. The first model to be studied
(Dimopoulos and Raby , 1983 ; Alb recht et al., 1982, 1983) was the
'geometric hierarchy model' (proposed by Witten ( 1 98 1b)), a
supersymmetric model which makes use of a fundamental energy scale of
abou t 10 1 2 GeV. This model provided acceptable density fluctuations and
much more than adequate inflation , but the reheating temperatu re of the
model was too low to allow for baryogenesis. Subsequent work has led to a
number of models that are believed to be completely successful. In a series of
papers (Ellis et al., 1982, 1983a, b ; Nanopoulos et al., 1983a, b; Nanopoulos
and Srednicki, 1983 ; Nanopoulos, Olive and Srednicki, 1983 ; Gelmini,
Nanopoulos and Olive, 1983; Olive, 1983 ; Linde, 1983e,f; Ellis et al., 1984;
Kounnas and Quiros, 198 5 ; Ellis et al. , 198 5 ; Gelmini, Kounnas and
Nanopoulos, 198 5 ; Enqvist et al., 1985a, b; Jensen and Olive, 1985, 1986), a
group loosely centered at CERN has developed a model called 'primordial
inflation', in which inflation is driven by a scalar field which breaks
supersymmetry near the Planck scale. Alternative approaches using
supersymmetry and/or supergravity have been developed by Ovrut and
Steinhardt ( 1983, 1984a, b, c, 1986; see also Albrecht and Steinhardt, 1983 ;
and Lindblom, Ovrut and Steinhardt, 1986) and by Holman, Ramond and
Ross ( 1984). Shafi and Vilenkin ( 1984) have shown that supersymmetry is
not necessary for inflation - they constructed a model bas ed on a non
minimal SU(5) theory that includes a weakly coupled gauge singlet
'inflation' field. Pi ( 1984) developed a variant of this model in which the role
of the inflaton is played by the axion field. Other particle physics models that
are believed to lead to successful inflationary cosmologies have been
constructed by Hung (1984), Gupta and Quinn (1984), and Shafi and
Stecker ( 1984).
At the least, the models mentioned above demonstrate that the problems
associated with inflation in the context of the minimal S U(5) GUT can be
overcome. However, we think it is fair to say that all of these theories appear
somewhat contrived. The ultimate goal of constructing a theory which is
consistent with inflation and which also provides an elegant solution to the
problems of particle physics has yet to be achieved.
12. 1 1 False vacuum bubbles and child universes

In this section we will briefly discuss the behavior of false vacuum bubbles -
regions of infla_tion that are surrounded by regions that do not inflate . This
issue is relevant to all theories of the early universe in which the conditions
before the inflationary phase transition were highly inhomogeneous. It is
also relevant to the intriguing question of whether it is possible in principle
to create an inflationary universe 'in the laboratory' (i .e. , by man-made
processes) . Furthermore, the geometry of false vacuum bubbles provides a
very interesting example of a highly non-Euclidean situation .
The mathematics of inhomogeneous spacetimes can be very complicated,
but a detailed calculation can be carried out for one idealized system : a
spherical region of false vacuum surrounded by an infinite region of true
vacuum, with a thin domain wall separating the two regions. This situation
has been studied by a number of authors (Sato et al., 198 1 , 1982; Kodama et
al., 198 1 , 1982; Sato , 198 1c; Maeda et al., 1982; Berezin, Kuzmin and
Tkachev, 1983, 1985 ; Aurilia et al., 1984, 1985; Blau , Guendelman and
Guth, 1986).
Although this idealized system is highly simplified , it nonetheless raises a
significant paradox . If the false vacuum region is sufficiently large, then an
observer deep within it would unambiguously expect to see inflation. An
observer watching the domain wall, however, would note that the false
vacuum region has negative pressure and is surrounded by the zero pressure
true vacuum, so the pressure force is inward . The spherical symmetry
implies that the metric in the true vacuum region has the usual
Schwarzschild form, so the gravitational field is not expected to oppose the
force of the pressure gradient. Thus, the second observer would not expect to
see inflation. The resolution of this paradox relies on a highly non-Euclidean
geometry, allowing the false vacuum region to inflate without moving
outward into the true vacuum region .
An example of a false vacuum bubble solution is illustrated in Fig. 12.7.
The figure shows a diagram of spacetime, with the angular coordinates e and
</> suppressed - light-like lines travel at 45°. To the right of the bubble wall
(shown as a heavy line with an arrow on it) the diagram represents a region
of Schwarzschild space, shown in Kruskal-Szekeres coordinates. The
behavior of the solution can be illustrated more intuitively by the diagrams
in Fig . 12.8, which show the spatial hypersurfaces at successive values of the
time coordinate used in Fig . 12.7. The diagrams are drawn by suppressing
one dimension of the hypersurface and embedding the resulting two
dimensional surface in a three-dimensional space so that the curvature can
be displayed. The labels (a)-(d) in Fig . 12.8 correspond to the horizontal
slices indicated in Fig . 12.7.
The resolution of the paradox is now apparent : the shape is distorted so
that a force driving the bubble wall from the true vacuum to the false
vacuum pushes the wall to larger values of the radius. Meanwhile the false
Fig . 12.7. A spacetime diagram of a false vacuum bubble solution. Angular

coordinates are suppressed , and lightlike lines travel at 45°. The bubble wall
is shown as a heavy line with an arrow on it. The true vacuum region is to the
right of the wall, where the metric describes part of Schwarzschild space,
shown in Kruskal-Szekeres coordinates. The false vacuum region is to the
left, where the metric describes part of de Sitter space. The horizontal lettered
lines indicate spacelike hypersurfaces to be illustrated in Fig. 12.8.
c
r= 0
non-singular
b
singular
592 S. K. Blau and A . H . Guth
vacuum region expands into a bulge which soon disconnects completely

from the original space, forming an isolated closed universe. Sato et al.
( 1982) have called this disconnecting region a 'child universe'. A black hole
singularity is left in the original spacetime, but for typical GUT parameters
it would evaporate in about 10- 14 s.
Figs. 12.7 and 12.8 illustrate one type of solution to the false vacuum
bubble equations of motion, but there are also solutions describing small
bubbles which are crushed by the pressure forces. There are also 'bounce'
solutions, describing bubbles which approach infinite size in the asymptotic
past, fall to a minimum size, and then grow without bound . At late times
these bounce solutions resemble the solution shown in Figs. 12.7 and 12.8.
Although the problem that has been solved is very idealized, we believe
that it contains the essential physics of more complicated inhomogeneous
spacetimes. The paradox discussed above will exist whenever an inflating
region is surrounded by a non-inflating region, and the qualitative behavior
of the system seems to be determined by the way in which this paradox is
Fig. 12.8. The evolution of a false vacuum bubble solution . Each lettered
diagram illustrates a spacelike hypersurface indicated in Fig. 1 2 . 7. The
diagrams are drawn by suppressing one dimension of the hypersurface and
embedding the resulting two-dimensional surface in a three-dimensional
space so that the curvature can be displayed. The false vacuum region is
indicated by shading. Note that diagram (d) shows a child universe detaching
from the original spacetime.
I "'v c I / ,g c I
(a) (b)
_
I ____._.
, _.,...
... /___,/
v
(c) (d)
resolved . Thus, we believe that if inflation occurred in an inhomogeneous

universe, then many isolated 'child universes' would have been ejected .
Furthermore, the analysis of Section 12.9 indicates that even if inflation
began homogeneously, child universes are likely to have been produced . To
see this, consider the field $1(x, t), smeared over a physical size x - 1 . In the
approximations of Section 12.9, this quantity obeyed a Gaussian
probability distribu tion, with a root-mean-square value which (for µ > 0)
increased exponentially with time. The probability density for ¢1 � o
therefore decreased exponentially with time, but for µ 2 � x 2 this decrease
was considerably slower than the exponential increase in the volume,
proportional to e 3 x1 . Thus, the expected number of horizon-sized volumes
with ¢1 � o increased with time. t Although these regions were presumably
non-spherical , it seems likely that they would have evolved into child
um verses.
The question of whether a false vacuum bubble, and hence an inflationary
universe, can in principle be produced in the laboratory has been
investigated by Farhi and Guth ( 1986). The issue concerns the initial
singularity shown in Fig . 12.7. Although an initial singularity is often
hypothesized to have been present at the big bang, there do not appear to be
any initial singularities available today. Thus, to produce an inflationary
universe in the laboratory, one must find some mechanism to modify the
early stages of this pictu re, so that the initial matter distribution is smooth. It
can be shown , however, that the singularity theorems of general relativity
require the initial singularity, provided that
(i) the false vacuum region inflates without bound ;
(ii) the energy-momentum tensor obeys the condition that
Tµ v w11 wv ::;.--
�o ( 1 1.)
for any null vector Wµ ; and
(iii) the false vacuum bubble is spherically symmetric.
The proof does not require that the machinery which produces the bubble be
spherically symmetric. Intuitively, the theorem means that the outward
velocity required for the false vacuum bubble to grow without bound is so
large that it can emerge only from an initial singularity . Although a rigorous
argument exists only for the limited case described above, it seems
reasonable to conjecture that the theorem applies even in the absence of
spherical symmetry (condition (iii)). Note, however, that this argument does
not exclude the possiblity of the creation of a false vacuum bubble from
t We thank Alex Vilenkin for calling our attention to this fact.
smooth initial conditions by a quantum tunneling process, since the

quantum mechanical energy-momentum tensor violates condition (ii)
above.
12. 12 Conclusion
As emphasized by Tu rner (1985), inflation is really a paradigm, and not a
specific theory. In particular, there are a number of variations of inflation
that appear to be capable of achieving the same objectives. Of these,
probably the best known is Linde's proposal of chaotic inflation (Linde,
1983b, c, d, 1986a, b, c; see also Goncharov and Linde, 1984a, b, c ;
Goncharov e t al., 1984; and Khlopov and Linde, 1984). Our primary reason
for omitting a discussion of chaotic inflation was the simple fact that Linde is
also contributing an article to this volume (Chapter 13). The key idea of
chaotic inflation is the observation that inflation can work for a scalar field
potential as simple as V( </>) = A</> 4 , provided that one makes some
assumptions about the initial conditions. Linde proposes that the scalar
field begins in a chaotic state, so that there are some regions in which the
value of </> is a few times larger than MP . These regions must exceed some
minimal size, which is estimated to be somewhere between 2H - 1 (Linde,
1984a) and 9H - 1 (Turner, 1985), where H denotes the Hubble constant.
Then </> rolls down the hill of the potential energy diagram, and a
straightforward calculation indicates that there is an adequate amount of
inflation. The Hubble 'constant' is not a constant in this case, but it is slowly
varying, so the expansion can be called 'quasi-exponential' . In order for the
density fluctuations to have the desired magnitude of (L\p/p ) I H � 10- 4 , the
coupling constant A. must be very small : A. � 10 - 1 3. This number is typical of
working inflationary models. Thus, chaotic inflation can greatly enlarge the
class of particle theories consistent with inflationary cosmology - the
plateau of the effective potential diagram is no longer required. It seems
difficult to assess the plausibility of the initial conditions required for chaotic
inflation, but the same can be said for the standard new inflationary
scenario. Either model appears to be much more plausible than the standard
cosmology (Section 12.2), and probably both models could benefit from
more work concerning this question.
Another variation of inflation is the Starobinsky model (Starobinsky,
1979, 1980), a model proposed in its original form several years before
inflation itself. In this model the energy density of the false vacuum is
supplied by the curved space quantum corrections to the renormalized
expectation value of the energy-momentum tensor of one or more scalar
fields in the theory (Dowker and Critchley, 1976). While the initial
conditions required for this model do not appear plausible at the level of
classical physics, the desired conditions can perhaps be achieved in models
in which the universe is created by a quantum fluctuation from 'nothing'.
For a summary of the model and its literature, see Linde ( 1984a) and
Vilenkin ( 1985c).
Other alternative inflationary models include induced gravity inflation
(Accetta, Zoller and Turner, 1985 ; Spokoiny, 1984), and inflation driven by
a scalar field which represents the radius of a compactified dimension (Shafi
and Wetterich, 1983, 1985). Both of these possibilities, along with chaotic
inflation, are summarized in Turner's review ( 1 985).
In conclusion , we want to say that the basic idea of inflation - the idea that
the universe went through a period during which it expanded exponentially
while trapped in a false vacuum - appears to us to be probably correct. It is a
very simple and natu ral idea in the context of spontaneously broken gauge
theories, and it seems to solve some very fundamental cosmological
problems. On the other hand , we clearly do not yet have the details straight.
In order to understand the density fluctuations, for example, we must at the
same time understand the details of particle physics at GUT energy scales.
We are presumably some distance from that goal .
Finally, we wish to emphasize that probably the most dramatic recent
development in cosmology is the realization that the universe may be
completely devoid of conserved quantum numbers. If so, then , even if we do
not understand the precise scenario, it becomes very plausible that our
observed universe emerged from nothing or from almost nothing. The
universe may indeed be the ultimate free lunch.
Acknowledgements
We would like to thank the many people with whom we have interacted over
the years, exchanging information , insights, opinions, and suggestions
concerning inflationary cosmology. A complete list is impossible to
construct, but the list should certainly include Larry Abbott, Jim Bardeen ,
M arc Davis, Lawrence Ford, Stephen Hawking, Lawrence Krauss, Andrei
Linde, Ian M oss, So-Young Pi, Bill Press, David Schramm, Marc Sher,
Gary Steigman, Paul Steinhardt, M ichael Turner, Henry Tye, Alex
Vilenkin , and Erick Weinberg. In addition, our understanding of density
fluctuations in the new inflationary univers� was aided by conversations
with Aleksei Starobinsky and Jemal Guven, and our understanding of false
vacuum bubbles was aided by the input of Edward Farhi, Jose Figueroa-
O'Farrill , Eduardo Guendelman, Werner Israel , Richard Matzner,

Malcolm Perry, Ian Redmount, Bill Unruh, Robert Wald , and Neil
Woodward . We also benefited from discussions about the stability of
de Sitter space with Bruce Allen , Joshua Frieman, David Garfinkle, Emil
Mottola, and Jennie Transchen, but we do not mean to imply that all of
these physicists agree with ou r point of view.
This paper is based in part on lectures delivered by A .H .G. at the First
Jerusalem Winter School of Theoretical Physics, and we would like to
acknowledge the warm hospitality of that meeting.
The work of S.K.B. is supported in part by funds provided by the Robert
A. Welch Foundation and in part by the US National Science Foundation
(NSF) under contract PHY 8304629. The work of A.H.G. is supported in
part by funds provided by the US Department of Energy (DOE) under
contract DE-AC02-76ER03069, and in part by the National Aeronautics
and Space Administration (NASA) under grant NAGW-553 .
Appendix
In this append-ix we will prove a mathematical property of the solutions to a
differential equation which was used in Section 12.8. The equation has the
form
(A . l)
where the dot denotes differentiation with respect to t, and where there exists
a t* such that w 2 (t) > 0 for t > t*. We will show that one can find two linearly
independent solutions u1 (t) and u 2(t) with the property that for t > t*, u 1 (t) is
monotonically increasing and l u 2 (t) I < constant x e - 3xi . Thu s , at large times
there is an essentially unique solution u 1 (t).
To prove the theorem, we first construct u1 (t) . This can be done by
choosing initial data at t t* , with u1 (t*) > 0 and u 1 (t* ) > 0. By writing eq.
=
(A. l) as
(A.2)
one can see immediately that u > 0 for all t > t* .
To construct u 2 (t), consider an arbitrary solution v(t) and construct the
Wronksian
= =
(A.3)
One can easily verify that W - 3x W, so one has W W0 e - 3xi. It follows
that
u21 i_ ()
�
d t U1 = w0 e 3 r '
- x (A.4)
which can be integrated to give
v(t) = c1 u1 (t) - W0 u1 (t)

oo -2-, dt'.
e - 3 xt'
J u 1 (t ) (A. 5)
Thus we can take
oo e - 3 xt'
U2 (t) = U1 (t)
The monotonicity of u 1 implies that for
J u 1 (t ) d t .
�'
t > t* ,
' (A. 6)
e - 3xr e - 3xr
0 < u 2 (t) < < . (A. 7)
3xu1 ( t) 3xu1 (t*)
References
Aaronson , M . , Mould , J., Huchra, J., Sullivan, W. T., Schommer, R . A. and Bothun,
G . D . ( 1980). Astrophys. J., 239, 12, 66-B l .
Abbott , L . F . ( 1 98 1). Nucl. Phys., B 185, 233.
Abbott, L. F., Farhi, E . and Wise, M. B. ( 1982). Phys. Lett., 1 17B, 29.
Abbott , L . and Pi, S .-Y. ( 1986). Inflationary Cosmology. World Scientific : Singapore .
Abbott , L. F. and Sikivie, P . ( 1983). Phys. Lett., 120B, 133.
Abbott, L. F . and Wise, M . B . ( 1984a). Phys. Lett., 135B, 279.
Abbott, L. F. and Wise, M . B. ( 1 984b). Astrophys. J., 282, L47 .
Abramowitz, M . and Stegun , I . A . ( 1 972). Handbook of Mathematical Functions with
Formulas, Graphs, and Mathematical Methods. Dover : New York .
Accetta, F., Zoller, D . and Turner, M . S . ( 1985). Phys. Rev., D31 , 3046.
Albrecht, A. and Brandenberger, R. ( 1985). Phys. Rev., D31 , 1225 .
Albrecht, A . , Brandenberger, R . and Matzner, R . ( 1985). Phys. Rev., D32, 1280.
Albrecht, A . , Dimopoulos, S., Fischler, W., Kolb, E. W . , Raby, S. and Steinhardt, P . J.
( 1 982). Proceedings of the 3rd Marcel Grossman Meeting on the Recent Developments
in General Relativity, ed . Hu Ning, p . 5 1 1 . North-Holland : Amsterdam.
Albrecht, A . , Dimopoulos, S . , Fischler, W . , Kolb, E. W . , Raby, S. and Steinhardt, P. J.
( 1983). Nucl. Phys., B229, 528 .
Albrecht, A . and Steinhardt, P . J . ( 1982). Phys. Rev. Lett., 48, 1220.
Albrecht, A. and Steinhardt, P. J. ( 1983). Phys. Lett. , 131B, 45 .
Albrecht, A . , Steinhardt, P . J . , Turner, M . S. and Wilczek, F. ( 1982). Phys. Rev. Lett.,
48, 1437.
Allen , B. ( 1986). The graviton propagator in de Sitter space, Tufts preprint TUTP 86-9.
Antoniadis, I . , Iliopoulos, J. and Tomaras, T. N. ( 1 986). Phys. Rev. Lett. , 56, 1 3 19.
Arya! , M ., Everett, A . E., Vilenkin, A . and Vachaspati, T. ( 1986). Phys. Rev., 034, 434.
Atkatz, D. and Pagels, H . ( 1982). Phys. Rev. , D25, 2065.
Aurilia, A., Denardo, G., Legovini, F. and Spallucci, E. ( 1984). Phys. Lett., 1478, 258.
Aurilia, A., Denardo, G., Legovini, F. and Spallucci, E. ( 1985). Nucl. Phys., 8252, 523.
Axenides, M., Brandenberger, R. and Turner, M. S. ( 1983). Phys. Lett., 1268, 178.
Bardeen, J . M . ( 1980). Phys. Rev., D22, 1882.
Bardeen, J . M . , Bond, J . R . , Kaiser, N. and Szalay, A . S. ( 1 986). The statistics of peaks
of Gaussian random fields. To be published in Astrophys. J.
Bardeen, J. M ., Steinhardt, P. J . and Turner, M . S. ( 1983). Phys. Rev., D28, 679.
Barrow, J. D. ( 1983). In The Very Early Universe (Proceedings of the Nuffield
Workshop), ed. G. W. Gibbons, S. W. Hawking and S . T. C. S iklos, p. 267.
Cambridge University Press: Cambridge.
Barrow, J. D. ( 1986). The deflationary universe : an instability of the de Sitter u niverse.
Sussex preprint.
Bennett, D. P. ( 1986) . Phys. Rev., D33, 872.
Berezin, V . A., Kuzmin , V . A . and Tkachev, I . I . ( 1983). Phys. Lett. , 1208, 9 1 .
Berezin, V . A . , Kuzmin , V . A . and Tkachev, I . I . ( 1985). I n Proceedings of 3rd Seminar
on Quantum Gravity, 1 984 , ed . M . A. Markov, V. A. Berezin and V. P. Frolov, p. 605.
World Scientific : S ingapore.
Bernard, C. ( 1974). Phys. Rev., D9, 33 13.
Billoire, A. and Tamvakis, K . ( 1982). Nucl. Phys., 8200, 329.
Blau, S . K . , Guendelman, E. I . and Guth, A. H . ( 1986). The dynamics of false vacuum
bubbles. Submitted to Phys. Rev., D .
Bludman, S. A . and Ruderman, M . A . ( 1977). Phys. Rev. Lett., 38, 255 .
Boesgaard, A. and Steigman, G . ( 1985). To be published in Ann . Rev. Astron. Astrophys.
Boucher, W. and. Gibbons, G. W. ( 1983) . In The Very Early Universe (Proceedings of
the NulTield Workshop); ed . G. W. Gibbons, S. W. Hawking and S. T. C. Siklos ,
p. 273 . Cambridge University Press : Cambridge.
Branch, D. ( 1979). Mon. Not. R. Astron. Soc., 186, 609.
Brandenberger, R. H. ( 1984). Nucl. Phys., 8245, 328.
Brandenberger, R . H . ( 1985). Rev. Mod. Phys., 57, 1 .
Brandenberger, R . , Kahn , R . and Press, W . H . ( 1983). Phys. Rev., D28, 1809.
Brandenberger, R . H . and Turok, N. ( 1986). Phys. Rev., D33, 2 182.
Brout, R . , Englert , F. and Spindel , P . ( 1979). Phys. Rev. Lett., 43, 4 17.
Bunch, T. S . and Davies, P . C. W. ( 1978). Proc. Roy. Soc. London, Ser., A360, 1 17.
Buras, A. J . , Ellis, J . , Gaillard, M . K . and Nanopoulos, D. V . ( 1978). Nucl. Phys., 8 1 35,
66.
Callan, C. G. and Coleman, S. ( 1977). Phys. Rev., D l6, 1762.
Chudnovsky, E. M., Field , G . B., Spergel , D . N. and Vilenkin, A . ( 1986).
Superconducting cosmic strings. Submitted to Phys. Rev., D .
Coleman , S . ( 1 975) . I n Laws of Hadronic Matter (Erice, 1973), ed . A . Zichichi , p . 1 39 .
Academic Press : New York .
Coleman, S . ( 1977) . Phys. Rev., D IS, 2929.
Coleman, S. ( 1979). In The Whys of Subnuclear Physics (Proceedings of the
International School of Subnuclear Physics, Ettore Majorana, Erice, 1977), ed.
A. Zichichi. Plenum : New York.
Coleman, S . and De Luccia, F. ( 1980). Phys. Rev. , D21 , 3305.
Coleman, S . and Weinberg, E. J . ( 1973) . Phys. Rev., D7, 1 888 .
Cook, G . P . and Mahanthappa, K . T. ( 1 982). Phys. Rev., D25, 1 1 54.
Dicke, R . H . and Peebles, P. J . E. ( 1979) . In General Relativity: An Einstein Centenary
Survey, ed . S . W . Hawking and W . Israel . Cambridge University Press : Cambridge.
Dicus , D . , Kolb , E. W. and Teplitz, V. ( 1977). Phys . Rev. Lett., 39, 168 .
Dimopoulos, S . and Raby, S . ( 1983) . Nucl. Phys., B219, 479.
Dine, M . and Fischler, W. ( 1983). Phys. Lett., 1 20B, 137.
Dolan, L . and Jackiw, R. ( 1974a). Phys. Rev., 09, 2904.
Dolan , L. and Jackiw, R. ( 1974b) . Phys. Rev., 09, 3320.
Dolgov, A . D. and Linde, A . D. ( 1982). Phys. Lett., 1 1 6B, 329.
Dowker, J. S. and Critchley, R. ( 1976). Phys. Rev., 0 13, 3224.
Einhorn, M . B. and Sato, K. ( 198 1). Nucl. Phys., B180 [FS2] , 385.
Ellis , J . , Enqvist, K . , Gelmini, G . , Kounnas, C., Masiero, A . , Nanopoulos, D . V . and
Smirnov, A. Yu . ( 1 984). Phys. Lett., 147B, 27 .
Ellis, J . , Enqvist, K . , Nanopoulos, D. V., Olive, K. A . and Srednicki, M. ( 1985). Phys.
Lett . , 152B, 175 [Erratum: Phys. Lett., 156B, 452 ( 1985)] .
Ellis, J . , Nanopoulos, D . V . , Olive, K . A. and Tamvakis, K . ( 1982). Phys. Lett. , 1 18B,
335.
Ellis, J . , Nanopoulos, D. V., Olive, K . A . and Tamvakis, K . ( 1983a). Nucl. Phys., B221 ,
524.
Ellis, J . , Nanopoulos, D. V., Olive, K. A. and Tamvakis, K. ( 1983b). Phys. Lett. , 120B,
33 1 .
Ellis, J . and Steigman, G . ( 1980). Phys. Lett., B89, 186 .
Enqvist, K . , Nanopoulos, D . V . , Quiros, M . and Kounnas, C. ( 1985a). Nucl. Phys.,
B262 , 538.
Enqvist, K., Nanopoulos, D . V . , Quiros, M. and Kounnas, C. ( 1985b). Nucl. Phys.,
B262 , 556.
Faber, S . M. and Gallagher, J. S . ( 1979). Ann. Rev. Astron. Astrophys., 17, 135.
Farhi, E . and Gu th , A . H. ( 1986). An obstacle to creating a universe in the laboratory,
unpublished .
Ford , L . H . ( 1985). Phys. Rev. , 031 , 7 10.
Frieman , J . A . and Will , C. M. ( 1982). Astrophys. J., 259, 437.
Gelmini, G. B., Kounnas, C. and Nanopoulos, D. V. ( 1985). Nucl. Phys., B250, 177.
Gelmini, G. B., Nanopoulos, D. V. and Olive, K . A . ( 1983). Phys. Lett., 131B, 53.
Georgi , H. and Glashow, S . L. ( 1974). Phys. Rev. Lett. , 32 , 438.
Georgi, H . , Quinn, H. R. and Weinberg, S . ( 1974). Phys. Rev. Lett. , 33, 45 1.
Gibbons, G . W. and Hawking, S . W. ( 1977). Phys. Rev., 0 15, 273 8.
Ginsparg , P . & Perry, M. J . ( 1983). Nucl. Phys., B222 , 245 .
Gliner, E. B . ( 1965) . Zh . Eksp. Teor. Fiz., 49, 542 (JETP Lett. , 22, 378 ( 1966)).
Gliner, E. B. ( 1970) . Dok/. Akad. Nauk SSSR, 192 , 771 (Sov. Phys. Doklady, 15, 559
( 1970)).
Gliner, E. B. and Dymnikova, I . G. ( 1975). Pis'ma Astron. Zh ., 1 , 7 (Sov. Astr. Lett., 1 ,
9 3 ( 1975)).
Goldman, T., Kolb., E. W. and Toussaint, D. ( 198 1). Phys. Rev., 023, 867.
Goncharov, A. S. and Linde, A . D. ( 1984a). Zh . Eksp. Teor. Fiz., 86, 1 594 (JETP, 59,
930 ( 1984)).
Goncharov, A. S. and Linde, A. D. ( 1984b) . Phys. Lett ., 139B, 27.
Goncharov, A. S. and Linde, A. D. ( 1984c). Class. Quantum Grav., 1, L 75.
Goncharov, A. S., Linde, A. D . and Vysotskii, M. I . ( 1984). Phys. Lett., 1478, 279.
Gott, J. R. ( 1982) . Nature, 295, 304.
Gupta, S . and Quinn, H . R. ( 1984). Phys. Rev., 029, 279 1 .
Guth, A . H. ( 198 1). Phys. Rev., 023, 347.
Guth, A . H . ( 1983a). In Asymptotic Realms of Physics: Essays in Honor of Francis E.
Low, ed. A. H. Guth, K . Huang and R. L. Jaffe. MIT Press : Cambridge,

Massachusetts.
Guth, A. H. ( 1983b). In Magnetic Monopoles (Proceedings of the NATO 'Advanced
Study Institute on Magnetic Monopoles, Wingspread, Wisconsin, 1 982), ed . R. A.
Carrigan and W. P . Trower. Plenum : New York .
Guth, A . H . and Pi, S .-Y . ( 1982). Phys. Rev. Lett. , 49, 1 1 10.
Guth, A . H . and Pi, S .-Y. ( 1985). Phys. Rev., 032 , 1899.
Guth, A. H . and Pi, S.-Y. ( 1986). In Inner Space/Outer Space: The Interface between
Cosmology and Particle Physics, ed. E. W. Kolb, M. S. Turner, D. Lindley, K . Olive
and D. Seckel, p. 345 . Chicago University Press : Chicago.
Guth, A. H . and Sher, M . ( 1983). Nature, 302, 505.
Guth , A. H . and Tye, S.-H . ( 1 980). Phys. Rev. Lett., 44, 63 1 [Erratum: Phys. Rev. Lett.,
44, 963 ( 1980)] .
Guth, A. H . and Weinberg, E. J . ( 198 1). Phys. Rev., D23, 876.
Guth, A . H . and Weinberg, E . J . ( 1983). Nucl. Phys., 821 2 , 32 1 .
Harrison, E . R. ( 1970) . Phys. Rev., D l , 2726.
Hartle, J . B . and Hawking, S. W. ( 1983). Phys. Rev., D28, 2960.
Hawking, S. W. ( 1982). Phys. Lett., 1 158, 295.
Hawking, S. W. ( 1984a). In Relativity, Groups and Topology II (Proceedings of the Les
Houches Summer School in Theoretical Physics, June 27-4 Aug., 1983), ed. B. S .
DeWitt and R . Stora, p . 3 3 3 . North Holland : Amsterdam.
Hawking, S. W. ( 1 984b). Nucl. Phys., 8239, 257.
Hawking, S . W. ( 1986). The density matrix of the universe. Cambridge University
preprint 86-09 1 8 .
Hawking, S. W . and Ellis, G. F . R. ( 1973). The Large Scale Structure of Space-Time.
Hawking, S. W. and Luttrell, J . C. ( 1984). Phys. Lett., 1438, 83.
Hawking, S . W. and Moss, I . G . ( 1982). Phys. Lett. , 1 108, 35.
Hawking, S . W. and Moss, I . G . ( 1983). Nucl. Phys., 8224, 1 80.
Hawking, S . W., Moss, I . G . and Stewart, J . M . ( 1982). Phys. Rev., D26, 268 1 .
Hawking, S . W . and Wu , Z . C . ( 1985). Phys. Lett., 1518, 1 5 .
Holman, R . , Ramond, P . and Ross, G . G . ( 1984). Phys. Lett., 1378, 343 .
Hung, P . Q . ( 1984). Phys. Rev., D30, 1637.
IAU ( 1986). Dark Matter in the Universe (Proceedings of IAU Symposium 1 17 ,
Princeton, New Jersey, June 1985). Reidel : Dordrecht .
Ipser, J . and Sikivie, P. ( 1983). Phys. Rev. Lett., 50, 925.
Jackiw, R . ( 1974). Phys. Rev., D9, 1686.
Jensen, L. G. and Olive, K . A . ( 1985). Phys. Lett., 1 598, 99.
Jensen, L . G. and Olive, K. A . ( 1986). Nucl. Phys., 8263, 73 1 .
Jensen, L . G. and Stein-Schabes, J . A . ( 1 986). A no-hair theorem fo r inhomogeneous
cosmologies, Fermilab-Pub-86/5 1-A, unpublished .
Kaiser, N. ( 1983) . Astrophys. J., 273, L l 7 .
Kaiser, N. ( 1986). I n Inner Space/Outer Space: The Interface between Cosmology and
Particle Physics, ed . E. W. Kolb, M. S. Turner, D. Lindley, K. Olive and D. Seckel .
Chicago University Press: Chicago.
Khlopov, M . Yu. and Linde, A. D. ( 1 984) . Phys. Lett., 1388, 265 .
Kibble, T. W . B . ( 1976) . J. Phys., A9, 1387.
Kibble, T. W . B . ( 1986). Phys. Rev., D33, 328.
Kirzhnits, D . A . and Linde, A . D . ( 1972) . Phys. Lett., 428, 47 1 .
Kodama , H . , Sasaki, M . and Sato, K . ( 1982) . Prog. Theor. Phys., 68, 1979.
Kodama , H . , Sasaki , M ., Sato, K. and Maeda, K. ( 198 1 ) . Prog . Theor. Phys., 66, 2052.
Kolb, E. W. and Turner, M . S. ( 1983). Ann. Rev. Nucl. Part. Sci., 33, 645 .
Kolb , E . W. and Wolfram, S . ( 1980) . Astrophys. J., 239, 428 .
Kounnas, C. and Quiros, M . ( 1985). Phys. Lett., 151B, 189.
Langacker, P. ( 198 1). Phys. Rep., 72C, 185.
Lazarides, G ., Shafi , Q. and Trower, W. P . ( 1982). Phys. Rev. Lett., 49, 1756.
Lindblom, P . R . , Ovrut, B . A. and Steinhardt, P. J . ( 1986) . Phys. Lett. , B172, 309.
Linde, A. D. ( 1974) . Zh. Eksp. Teor. Fiz., 19, 320 (JETP Lett., 19, 183 ( 1974)) .
Linde, A . D . ( 1979) . Rep. Prog . Phys., 42, 389.
Linde, A. D . ( 1982a). Phys. Lett., 108B, 389.
Linde, A. D. ( 1982b) . Phys. Lett., 1 14B, 43 1 .
Linde, A . D . ( 1982c) . Phys. Lett., 1 16B, 335.
Linde, A . D . ( 1983a) . In The Very Early Universe (Proceedings of the Nuffield
Workshop), ed . G. W. Gibbons, S . W. Hawking and S. T. C. Siklos, p. 205 .
Linde, A. D. ( 1983b) . Zh. Eksp. Teor. Fiz., 38, 149 (JETP Lett., 38, 176) .
Linde, A . D . ( 1 983c) . Phys. Lett., 129B, 177.
Linde, A . D. ( 1 983d) . Phys. Lett., 131B, 330.
Linde, A . D . ( 1983e). Phys. Lett. , 132B , 3 17.
Linde, A . D . ( 1983!). Zh. Eksp. Teor. Fiz., 37, 606 (JETP Lett., 37, 724 ( 1983)).
Linde, A . D. ( 1 984a). Rep. Prog. Phys., 47, 925 .
Linde, A . D . ( 1984b). Nuovo Cim. Lett. , 39, 40 1 .
Linde, A . D . ( 1984c) . Zh. Eksp. Teor. Fiz., 87, 369 (JE TP Lett., 60(2), 2 1 1 ( 1984)).
Linde, A . D . ( 1984d) . Zh. Eksp. Teor. Fiz., 40, 496 (JETP Lett., 40, 1333 ( 1984)).
Linde, A . D . ( 1985). Phys. Lett., 158B , 375.
Linde, A . D. ( l986a) . Mod. Phys. Lett., Al, 8 1 .
Linde, A . D . ( 1986b) . Eternally existing, self-reproducing inflationary universe. To be
published in Proceedings of Nobel Symposium on Unification of Fundamental
Interactions.
Linde, A. D. ( 1986c) . Eternally existing self-reproducing chaotic inflationary universe.
Phys. Lett., 1 75B, 395.
Lindley, D . ( 1985). The inflationary universe : a brief history. Unpublished.
Maeda, K . , Sato , K . , Sasaki , M . and Kodama , H . ( 1982) . Phys. Lett. , 108B, 98.
Mazenko, G . F., Unruh, W. G . and Wald, R. M. ( 1985). Phys. Rev., D31 , 273.
Mazur, P. and M ottola, E. ( 1986) . Spontaneous breaking of de Sitter symmetry by
radiative effects. Santa Barbara Inst. Theor. Phys. preprint NSF-ITP-85- 153.
Misner, C. W . , Thorne, K . S . and Wheeler, J . A. ( 1973). Gravitation. Freeman : San
Francisco .
Moss, I . G . ( 1983). Phys. Lett., 1 28B, 385.
Moss, I . G . ( 1984). Nucl. Phys., B238, 436.
Moss, I. G. and Wright, W. A . ( 1984). Phys. Rev., D29, 1067.
M ottola, E. ( 1985). Phys. Rev., D31 , 754.
Mottola, E . ( 1986). Phys. Rev., D33, 16 16.
M yhrvold, N . P . ( 1983a) . Phys. Rev., D28, 2439.
Myhrvold, N . P. ( 1983b). Phys. Lett., 132B, 308 .
Nanopoulos, D. V., Olive, K . A . and Srednicki, M . ( 1983). Phys. Lett., 127B, 30.
Nanopoulos, D. V., Olive, K. A., Srednicki , M . and Tamvakis, K. ( 1983a). Phys. Lett.,
123B, 4 1 .
Nanopoulos, D. V., Olive, K . A., Srednicki, M. and Tamvakis, K. ( 1983b). Phys. Lett.,
124B, 1 7 1 .
Nanopoulos, D . V . and Srednicki, M . ( 1983). Phys. Lett., 133B, 287.
Olive, K . A. ( 1983). In Galaxies and the Early Universe (Proceedings of the 18th
Rencontre de Moriond, Astrophysics Meeting), ed. J . Audouze and
J. Tran Thanh Van, p. 3. Reidel : Dordrecht.
Olson, D. W . ( 1976). Phys. Rev., D 14, 327.
Ostriker, J. P ., Thompson, C. and Witten, E. ( 1986) . Cosmological effects of
superconducting strings. Submitted to Phys. Lett.
Ovrut, B. A. and Steinhardt, P. J. ( 1983). Phys. Lett., 1 33B, 1 6 1 .
Ovrut, B . A . and Steinhardt, P . J . ( 1984a) . Phys. Rev. Lett., 53, 732.
Ovrut, B . A. and Steinhardt, P . J. ( 1984b) . Phys. Rev., D30, 206 1 .
Ovrut, B . A . and Steinhardt, P . J . ( 1984c) . Phys. Lett. , 147B, 263 .
Ovrut, B . A. and Steinhardt, P. J . ( 1986). In Inner Space/Outer Space: The Interface
between Cosmology and Particle Physics, ed. E. W . Kolb, M. S. Turner, D. Lindley,
K. Olive and D. Seckel . Chicago University Press : Chicago.
Peebles , P. J . E. ( 1979). Astron . J., 84 , 730.
Peebles, P. J . E. ( 1984) . Astrophys. J., 284 , 439 .
Pi, S .-Y . ( 1984) . Phys. Rev. Lett., 52, 1725.
Preskill, J . P . ( 1979). Phys. Rev. Lett., 43, 1365.
Preskill , J., Wise, M. B . and Wilczek , F. ( 1983). Phys. Lett. , 1 20B, 127.
Press, W . H. and Vishniac, E. T. ( 1980) . Astrophys. J., 239, 1 .
Rindler, W . ( 1956) . Mon . Not. R . Astron. Soc., 1 16 , 663 .
Sandage, A . and Tammam;1, G . A . ( 1 976) . Astrophys. J., 2 1 0 , 7 .
Sasaki, M . and Sato, K . ( 1982). Prag. Theor. Phys., 68 , 1979.
Sato , K . ( 1 98 1a). Phys. Lett., 99B, 66.
Sato, K . ( 198 1b). Mon . Not. R. Astron. Soc ., 1 95, 467 .
Sato, K . ( 198 lc). Prag . Theor. Phys., 66, 2287.
Sato , K., Kodama, H . , Sasaki, M . and Maeda , K . ( 1982) . Phys. Lett., 1 08B, 103 .
Sato, K . , Sasaki, M . , Kodama, H . and Maeda, K . ( 198 1 ) . Prag . Theor. Phys., 65 , 1443.
Scherrer, R. J. and Frieman, J . A. ( 1 986) . Phys. Rev., D33, 3556.
Schramm, D . N . and Steigman, G. ( 198 1). Astrophys. J., 243, 1 .
Seckel, D . and Turner, M . S . ( 1985). Phys. Rev., D32, 3 17 8 .
Shafi, Q . and Stecker, F. W . ( 1 984) . Phys. Rev. Lett., 53, 1 292.
Shafi, Q . and Vilenkin, A. ( 1 984) . Phys. Rev. Lett., 52, 69 1 .
Shafi, Q. and Wetterich, C . ( 1983). Phys. Lett. , 129B, 387.
Shafi, Q. and Wetterich, C. ( 1985). Phys. Lett. , 152B, 5 1 .
Sher, M . ( 198 1). Phys. Rev., D24, 1699.
Silk , J . and Turner, M. S. ( 1986) . Double inflation. Submitted to Phys. Rev., D .
Spokoiny , B . L . ( 1984). Phys. Lett. , 147B, 39 .
Starobinsky, A . A . ( 1 979). Zh. Eksp. Tear. Fiz., 30, 7 1 9 (JETP Lett. , 30, 682 ( 1979)).
Starobinsky, A. A. ( 1980) . Phys. Lett., 91B, 99 .
Starobinsky, A . A . ( 1982). Phys. Lett., 1 17B, 175.
Steigman, G . ( 1983). In Unification of the Fundamental Particle Interactions II
(Proceedings of the Europhysics Study Conference, Erice, Italy, Oct. 6- 14, 198 1 ) , ed .
J . Ellis and S . Ferrara . Plenum : New York .
Steinhardt, P . J . ( 198 la). Phys. Rev., D24, 842.
Steinhardt, P . J . ( 198 lb). Nucl. Phys., B190, 583.
Steinhardt, P . J . ( 1986). In High Energy Physics, 1 985 (Proceedings of the Yale
Theoretical Advanced Study I nstitute), ed . M . J. Bowick and F. Giirsey, Vol . 2, p. 567.

World Scientific : Singapore.
Steinhardt, P. J. and Turner, M . S. ( 1983). Phys. Lett., 129B, 5 1 .
Steinhardt, P . J . and Turner, M . S. ( 1984) . Phys. Rev., D29, 2 162.
Traschen, J. and Hill , C. T. ( 1986). Phys. Rev., D33, 3 5 19.
Tryon , E. P. ( 1973). Nature, 246, 396.
Turner, M . S. ( 1985). In Proceedings of the Cargese School on Fundamental Physics and
Cosmology, ed. J . Audouze and J. Tran Thanh Van. Editions Frontiers : Gif-Sur
Yvette.
Turner, M . S. ( 1986). Phys. Rev., D33, 889 .
Tu rner, M . S., Steigman , G. and Krauss, L. M . ( 1 984). Phys. Rev. Lett., 52, 2090.
Turner, M . S . , Wilczek, F. and Zee, A . ( 1983). Phys. Lett. , 1 25B, 35 [Erratum: Phys.
Lett. , 125B, 5 19 ( 1983)] .
Turok, N . ( 1986). Recent developments in the cosmic string theory of galaxy formation.
Lecture presented at the I nternational Symposium 'Particles and the Universe' at the
Aristotle Univ. of Thessaloniki, June 1985.
Turok, N . and Brandenberger, R. H . ( 1986). Phys. Rev., D33, 2 1 75 .
Vaucouleurs , G . d e and Bollinger, G . ( 1 979) . Astrophys. J., 233, 433 .
Vilenkin , A . ( 1982). Phys. Lett ., 1 17B, 25.
Vilenkin, A. ( 1983a). Phys. Rev., D27, 2848.
Vilenkin , A. ( 1983b). Nucl. Phys., B226, 527.
Vilenkin, A . ( 1983c). Nucl. Phys., B226, 504.
Vilenkin , A. ( 1984). Phys. Rev., 030, 509.
Vilenkin , A. (1985a). Phys. Rep., 121 , 263 .
Vilenkin, A. ( 1985b). Nucl. Phys., B252, 14 1 .
Vilenkin , A. ( 1985c). Phys. Rev., D32, 25 1 1 .
Vilenkin , A . and Ford, L . H . ( 1982) . Phys. Rev. , 026, 123 1 .
Vittorio, N ., Juszkiewica, R . and Davis, M . ( 1986). Large scale velocity fields as a test of
cosmological models. U npublished.
Vittorio , N. and Silk, J. ( 1985). Astrophys. J., 293, L l .
Vittorio, N . and Turner, M . S . ( 1986) . The large-scale peculiar velocity field in flat
models of the universe. Submitted to Astrophys. J.
Wald, R . M . ( 1983). Phys. Rev., D28, 2 1 1 8 .
Weinberg, S . ( 1972). Gravitation and Cosmology. Wiley : New York.
Weinberg , S . ( 1973). Phys. Rev ., D7, 2887 .
Weinberg, S . ( 1974). Phys. Rev., D9, 3357.
Wilkinson, D . T. ( 1986). In Inner Space/Outer Space: The Interface between Cosmology
and Particle Physics, ed. E . W. Kolb, M . S . Turner, D. Lindley, K . Olive and
D. Seckel, p. 126. Chicago University Pres s : Chicago.
Witten, E. ( 198 1a). Commun. Math . Phys., 80, 38 1 .
Witten, E . ( 1 98 1b). Phys. Lett., 105B, 267.
Yang, J., Turner, M . S . , Steigman, G., Schramm , D. N. and Olive, K. ( 1984).
Astrophys. J., 281 , 493 .
Yoshimura, M . (198 1). In Grand Unified Theories and Related Topics (Proceedings of the
4th Kyoto Summer Institute), ed. M . Konuma and T. Maskawa. World Scientific :
Singapore.
'
Zeldovich, Ya. B. ( 1 968). Usp. Fiz. Nauk., 95, 209 (Sov. Phys. Uspekhi, 1 1 , 38 1 ( 1968)).
Zeldovich, Ya. B. (1972) . Mon. Not. R. Astron. Soc., 160, l P .
Zeldovich, Ya. B . and Khlopov, M . Y . ( 1978). Phys. Lett., B79, 239.
13
Inflation and quantum cosmology
ANDREI LIN D E
13. 1 Introduction
One of the most popular models for the evolution of the universe developed
at the beginning of the 1980s is the inflationary universe scenario. This
scenario in its present form (see e.g. Linde, 1 984c, 1986c; Blau and Guth, this
volume, Ch. 1 2) makes it possible to give answers to about ten different
problems relating both to cosmology and to elementary particle physics. It
then becomes possible to understand why the observable part of the
universe, with a size 1 028 cm, is so flat, homogeneous and isotropic, why
,....,
we do not see any primordial monopoles, what is the origin of galaxies, etc.
No alternative theory that can solve all these problems has been suggested
so far. Therefore it seems plausible that something like inflation actually did
occur in the very early stages of the evolution of the universe.
Historically, there were many different versions of the inflationary
universe scenario. For example, about 20 years ago it was argued (Gliner,
1965) that the superdense baryonic matter should have the vacuum-like
equation of state, p . - p, and its energy-momentum tensor should have the
form 'I'µ v ,...., gwo1A . This would lead to exponential expansion of the universe in
the very early stages of its evolution (Gliner, 1970; Gliner and Dymnicova,
1975; Gurevich, 1975). At present, however, it seems that the equation of
state of superdense baryonic matter should be not p = - p but p = !p .
The next stage i n the development of the inflationary universe scenario i s
associated with the Starobinsky model (Starobinsky, 1979, 1980). He noted
that the exponentially expanding Friedmann universe (de Sitter space) is an
unstable solution of the Einstein equations with quantum corrections
(Dowker and Critchley, 1976), which after the development of the instability
transforms into the hot Friedmann universe. The main pu rpose of this
model was to solve the singularity problem, which proved to be too
Inflation and quantum cosmology 605
complicated. M oreover, density perturbations, generated during the

exponential expansion of the universe in the original version of this model,
typically are much greater than the desirable value bp/p "' 10 - 4 (M ukhanov
and Chibisov , 198 1 , 1982).
Another line of attack was related to the unified models of weak, strong
and electromagnetic interactions. As shown in Linde ( 1974), Veltman ( 1974,
1975) and Dreitlein (1974), the energy-momentum tensor of a constant
scalar field </>, which plays an important role in these theories, is given by
'F,i v (<f> ) "'g µ v V(<f> ) , where V(<f>) is the effective potential of this field. The value
of the field </> at the minimum of V( </>) depends on the value of the
=
temperature T in the very early universe. Typically, the minimum of V(<f>) at
large T is displaced to </> 0, and with decrease in T, a phase transition to
some state = 0, the
total entropy of the universe after the phase transition may increase
considerably (Linde, 1979). As pointed out by Guth ( 198 1), this effect can
help to solve the horizon problem, the flatness problem and the primordial
monopole problem. His model (now called the old inflationary universe
scenario) was very simple and attractive. It was based on four important
assumptions.
( 1) Initially tpe universe was in a symmetric state </> 0 due to high
temperature effects.
=
(2) After the universe expands the field </> becomes trapped at a local
minimum of V ( <f> ) near <f> = O. With a decrease of temperature the total
energy-momentum tensor of matter becomes equal to 'F,i v gµv V(O).
The universe in such a state expands exponentially.
=
(3) The stage of exponential expansion (inflation) finishes at the moment of
=
the phase transition to the stable state </> </> 0 =t- 0.
(4) The phase transition occurs due to formation of bubbles with </> </> 0 . =
The process of reheating the universe occurs due to bubble wall
collisions.
Unfortunately, in this scenario the universe becomes extremely
inhomogeneous after reheating (Guth, 198 1 ; Hawking and Moss, 1982;
Guth and Weinberg, 1983). This problem was solved in the context of the
new inflationary universe scenario (Linde, 1982a-d ; Albrecht and
Steinhardt, 1982). In the new scenario the last two assumptions mentioned
above were abandoned. This scenario is still very popular. In my opinion,
however, this scenario also is not perfect. It can be realised only in
some theories with rather unnatural potentials V( </>) that are extremely flat
606 A. Linde
=
near </> 0 and are sufficiently curved near the global minimum of V(</> ) .
Density perturbations produced in this scenario are sufficiently small only if
V(O)/M� "' 10- 1 0 and if the field </> interacts extremely weakly with all other
fields (e.g. A. -- 10 - 1 2 for V(</>) -- V(O) - -!A. </> 4). The first condition means that
inflation in this scenario starts very late, which makes it impossible to solve
the horizon problem in this scenario if the universe is closed : the universe
typically collapses before the beginning of inflation (Linde, 1984c ). The
=
second condition implies that thermal effects typically cannot raise the field
</> to the top of the effective potential at </> 0, and in such a case inflation
does not occur at all (Linde, 1984c, 1985a, b). Consequently, despite many
efforts, no realistic versions of the new inflationary universe scenario have
been suggested so far.
The only scenario in which these problems do not arise is the chaotic
inflationary scenario (Linde, 1983b, 1984a, b, 1985a, b). The main idea of
this scenario is based on the investigation of the possibility of inflation in a
universe filled with some non-equilibrium initial distribution of the field </>,
without making any ad hoc assumptions (1)-(4) concerning thermal
equilibrium, supercooling, etc. This scenario is a most general one, and,
surprisingly enough, it works perfectly well in a wide class of theories with
fairly natural effective potentials V(</> ) . Therefore in the present paper we
shall consider only this version of the inflationary universe scenario.
The concrete models of inflation are modified with each new development
of the underlying elementary particle theory. However, some basic features
of these models remain intact. Many conceptual problems of the
inflationary cosmology have now been solved . But two problems are still
widely discussed at present.
The first problem is connected with the initial conditions in the early
universe. This problem is also related to the singularity problem. The main
part of the problem is not the existence of singularities in the universe, but
the statement (or the common belief) that the universe does not exist
eternally and that there exists 'some time at which there is no spacetime at
all'. What is the origin of the universe? Was it created in a singular state or
has it appeared due to a quantum ju mp 'from nothing'? Which initial
conditions in the new-born universe are most natural ?
Different cosmologists give different answers to these questions. Some of
these answers are based on a phenomenological description of the universe
soon after its formation (Linde, 1983b, 1985a, b). Another approach is based
on the investigation of the· wave function of the universe {DeWitt, 1967;
Wheeler, 1968). The choice of a particular wave function in this approach is
Injlation and quantum cosmology 607
crucially author-dependent . One of the most interesting suggestions was

given by Hartle and Hawking ( 1983). Later Hawking ( 1984) and Hawking
and Page ( 1986) have studied a possibility of realisation of the chaotic
inflation scenario in quantum cosmology with the help of the Hartle
Ha wking wave function of the universe. An alternative choice of the wave
function was suggested by some other authors (Linde, 1984a, b; Zeldovich
and Starobinsky, 1984; Rubakov, 1984; Vilenkin, 1983). These two
approaches lead to different answers to the question of which initial
conditions were realised in the very early stages of the evolution of the
universe. This is one of the main problems to be discussed in the present
paper.
Another problem is related to the uniqueness of the universe. The essence
of this problem was formulated by Einstein in his talk with E. Straus: 'What
I am really interested in is whether God could create the world differently.'
The answer to this question in the context of the inflationary cosmology
appears to be rather unexpected . Namely, the local structure of the universe
is determined by inflation, which occurs at the classical level. The universe
after inflation becomes locally flat , homogeneous and isotropic. However,
its global structure is determined by quantum effects. It proves that the large
scale quantum fluctuations of the scalar field </J generated in the chaotic
inflation scenario lead to an infinite process of creation and self
reproduction of inflationary parts of the universe (Linde, 1986a, b, c). In this
scenario the evolution of the inflationary universe has no end and may have
no beginning. As a result, the universe becomes divided into many different
domains (mini-universes) of exponentially large size, inside which all possible
(metastable) vacuum states are realised. One may say therefore that not only
could God create the universe differently, but in His wisdom He created a
universe which has been unceasingly producing different universes of all
possible types.
We shall start our presentation with a discussion of the simplest version of
the chaotic inflationary scenario . However, most of the qualitative results
that will be obtained are essentially model-independent.
1i2 Chaotic inflation

Let us consider the simplest model of inflation based on the theory of a
massive non-interacting scalar field </J with trye Lagrangian
M P2 1 m2 (2. 1)
!f' = _
R + - aµ ¢ aµ<P - - </J 2 •
16n 2 2
_
608 A. Linde
Here M; 2 = G is the gravitational constant, M P - 10 1 9 Ge V is the Planck

mass, R is the curvature scalar, m is the mass of the scalar field </J, m <:g_ M P · If
the classical field <P is sufficiently homogeneous in some domain of the
universe (see below), then its behaviour inside this domain is governed by
equations
(/j + 3H c/J = -d V/d</J , (2.2)
k Sn 1 (2.3)
H2 + = b </J 2 + V(</J ).).
•
a2 3M ffi :
Here V(</J) is the effective potential of the field <P (in our case V(</J) = !m2</J2),
H = a/a, a(t) is the scale factor of the locally Friedmannian universe (inside
the domain under consideration), k = + 1, - 1 , or 0 for a closed, open or flat
universe, respectively. If the field <P initially is sufficiently large (</J � Mp),
then the functions </J(t) and a(t) rapidly approach the asymptotic regime
mMp
</J(t) = <Po - t, (2.4)
2(3n) 1 1 2
(2.5)
According to (2.4) and (2.5), during a time r - <fJ/mM P the value of the field <P
remains almost unchanged and the universe expands quasi-exponentially :
a(t + dt) - a(t) exp(H dt) (2.6)
for dt � r = </J/mM P . Here
2n1 12 m</J
H= · (2.7)
73 Mp '
H � r - 1 for </J � Mp .
The regime of quasi-exponential expansion (inflation) occurs for <P �!M P .
For <P � !M P the field <P oscillates rapidly, and if this field interacts with other
matter fields (which are not written explicitly in (2. 1)), its potential energy
V(</J -!MP ) - m 2 Mffi is transformed into heat. The reheating temperature TR
may be of the order (mMp)112 or somewhat smaller, depending on the
strength of the interaction of the field <P with other fields. It is important that
TR does not depend on the initial value <Po of the field </J. The only parameter
which depends on <Po is the scale factor a(t), which grows exp((2n/M ffi )</J6)
times during inflation.
If, as is usually assumed, a classical description of the universe becomes
possible only when the energy-momentum tensor of matter becomes
smaller than M� , then at this moment aµ <jJ a µ <jJ � M� and V(</J) � M� .
Therefore the only constraint on the initial amplitude of the field 2 � M; . This gives a typical initial value of the field oµ </> � V(</>) - M; inside this
universe, then very soon oµ </> o µ </> becomes much smaller than V(</> ) , the
evolution of the universe becomes describable by eqs. (2.2)-(2.7), and after
inflation the total size of the universe becomes larger than
l - M ; 1 exp
(M2ni </>�) - Mi 1 exp (2nm2M P2 ) . (2.9)
For m - 10- 5Mp (see the next section),

1 - M; 1 exp( 10 1 1 ) ;(; 10 1 0 10 cm, (2. 10)
which is much greater than the size of the observable part of the universe
"' 102 8 cm.
After such a large inflation the term k/a 2 in (2.3) becomes negligibly small
compared with H 2 , which means that the universe becomes flat and its
geometry locally Euclidean. For similar reasons the universe becomes
locally homogeneous and isotropic. The density of all 'undesirable' objects
(monopoles, domain walls, gravitinos) created before or during inflation
becomes exponentially small, and they never appear again if the reheating
temperature TR is not too large.
We should like to emphasise that for a realisation of this scenario it is
sufficient that initially oµ </> oµ </> � V(</>) - M; in a domain of a smallest
possible size O(Mp- 1 ). Since oµ </> oµ <J> � M: , V(</>) � M: in any classical
spacetime, the above-mentioned condition is quite natural . Note that,
despite some recent claims (see, for instance, Turner, 1985; Enquist et al.,
1986), there is no need for the field )) and to be uniform on scales much larger than M; 1 • For a
more detailed discussion of initial conditions which are necessary for
inflation see Linde ( 1985a, b).
This is the general scheme of inflation as it can be understood at the
classical level . One may wonder whether one can reach similar conclusions
concerning initial conditions in the context of quantum cosmology, and
whether quantum fluctuations may lead to some modification of the
scenario discussed in this section.
610 A . Linde
13.3 Inflation and the wave function of the universe

One of the most ambitious approaches to cosmology is based on the
investigation of the Wheeler-DeWitt equation for the wave function 'I' of
the universe (Wheeler, 1968 ; DeWitt , 1967). However, this equation has
many different solutions, and a priori it is not quite clear which of these
solutions describes our universe.
A very interesting idea was suggested by Hartle and Hawking ( 1983).
According to their work (Hartle and Hawking, 1983), the wave function of
the ground state of the universe with a scale factor a filled with a scalar field
<P in the semi-classical approximation is given by
'1'0 (a, </J) ,.., exp( - S E (a, ¢)). (3 . 1)
Here Sda, ¢) is the Euclidean action corresponding to the Euclidean
solutions of the Lagrange equation for a('r) and </J('r) with the boundary
conditions a(O) = a, </J(O) = </J. The reason for choosing this particular
solution of the Wheeler-DeWitt equation was explained as follows. Let us
consider the Green's function of a particle which moves from the point (0, t')
to the point (x, t) :
(x ,O ! O, t' ) = 2: 'l'11 (x) 'l'11 (0) exp(i E) )

II
f
= dx(t) exp(i S(x(t)), (3.2)
where '1'11 is a complete set of energy eigenstates corresponding to the
energies E11 � 0. To obtain an expression for the ground-state wave function
'I' 0 (x), one should make a rotation t -+ - i -r and take the limit as -r' -+ - oc .
In the summation (3.2) only the term n = O with E 0 = 0 survives, and the
integral transforms into J dx(-r) exp( - Sdx(-r))) . Hartle and Hawking have
argued that the generalisation of this result to the case of interest in the semi
classical approximation would yield eq. (3. 1).
The gravitational action corresponding to the Euclidean section S 4 of
de Sitter space dS4 with a(-r) = H - 1 (</J) cos H-r is negative,
'
f [( )
SE (a ,,;..'P ) = - � d 11
2
da 2
dry
A
3
J
- a2 + a 4 . 3 nM� = - 3Mi
2 16 V( <P)
'
(3.3)
Here 1J is the conformal time, 11 = J dt/a(t), A = 8 n V/M� . Therefore,
according to (3. 1),
'1'0 (a, ¢) - exp( - SE(a, ¢)) - exp( 3MP4
16 V(</J)
. ) (3 .4)
Inflation and quantum cosmology 61 1
This means that the probability P of finding the universe in the state with
¢ = const, a = H - 1 (¢) = (3M�/8n V(¢)) 1 1 2 is given by
( )
P(¢) ,...., l\J'ol2 ,...., exp
3M4
8 V(;)
. (3. 5)
This expression has a very sharp maximum as V(¢) -+ 0. Therefore the
probability of finding the universe in a state with a large field ¢ and having a
long stage of inflation becomes strongly diminished. One can argue, of
course, that at V(¢) � M� the function P(¢) becomes constant, and if one
introduces some cut-off at V(<f>) -+ 0 and integrates over all values of the field
-
¢ from OC' to + oc , then the probability of finding the universe with large ¢
(with V(¢) � M�) becomes large (Hawking and Page, 1986). However, this
would break the rule according to which one should try not to appeal to the
processes which occur at densities much greater than the Planck
density M� .
There exists an alternative choice of the wave function of the universe. It
can be argued that the analogy between the standard theory (3.2) and the
,
gravitational theory (3.3) is incomplete. Indeed, there is an overall minus
sign in the expression for SE (a ¢) (3 .3), which indicates that the
gravitational 'energy' associated with the scale factor a is negative. (This is
related to the well-known fact that the total energy of a closed universe is
zero, being a sum of the positive energy of matter and the negative energy of
the scale factor a.) In such a case, to obtain \JI 0 from (3 .2) one should rotate t
-
not to ii' but to + ir ,t which leads to (Linde, 1984)
,
-
P - l 'P 0 (a , </> ll 2 - exp( 2 I S, (a, </> liJ - exp ( -:�;i} (3 .6)
Actually, this result is valid only if the evolution of the field ¢ is very slow, so
that this field acts only as a cosmological constant A (¢) = 8n V( ¢)/M� in
(3.3). Fortunately, this is indeed the case during inflation.
Later the same result was obtained by another method, devised by
Zeldovich and Starobinsky ( 1984), Rubakov ( 1984) and Vilenkin ( 1983).
-
This result can be interpreted as a probability of quantum tunnelling of the
universe from a = O (from 'nothing') to a = H 1 (¢). In complete agreement
with the results of the previous section, eq. (3 .6) predicts that a typical initial
value of the field ¢ is given by V(¢) ;...,, Mi (if one does not speculate about the
possibility that V(¢) � Mi), which leads to a very long stage of inflation.
It must be said that there is no rigorous proof of either eq. (3. 1) or eq. (3.6),
t In our opinion this does not lead to a change of sign of the gravitational constant, as
claimed by Hawking (this volume, Ch. 14). In any case, the usual rule of rotation does
not give '1' 0 if En < O .
612 A . Linde
and the physical meaning of creation of everything from 'nothing' is far from
clear. Therefore a deeper understanding of the physical processes in the
inflationary universe is necessary in order to investigate the wave function of
the universe 'I' 0(a, <P) and to suggest a correct interpretation of this wave
function. With this purpose we shall try to investigate the global structure of
the inflationary universe, and go beyond the minisuperspace approach used
in the derivation of eqs. (3. 1) and (3.6). I
1 3.4 Quantum fluctuations in the inflationary universe

As we have already mentioned, inflation makes the universe locally
homogeneous and isotropic. However, inflation also leads to the creation of
long-wave fluctuations bp(x) of the field <P (Vilenkin and Ford, 1982 ; Linde,
1982c ; Starobinsky, 1982). Fluctuations generated during a time �t H - 1
in our theory form an inhomogeneous distribution of the classical field b</J(x)
=
with a time-independent amplitude
(4. 1 )
and with initial wavelength �I - H - 1 . Later, their wavelength grows expo
nentially as a(i), eq. (2.5), and the field <P + b</J(x) becomes almost exactly
homogeneous, slowly decreasing according to eq. (2.4). However, at the
same time new perturbations of the field <P are generated, and so on. This
process looks like a Brownian motion of the field </J. Inhomogeneities of the
resulting distribution of the field <P lead to density perturbations bp(x),
which grow very slowly (logarithmically) as their wavelengths grow. On a
galactic scale bp/p ,..., lO(m/MP ), which at m ,..., 10 - 5 M P ,..., 10 1 4 Ge V, gives the
desirable value bp/p ,..., 10 - 4 necessary for galaxy formation (Mukhanov and
Chibisov, 198 1 , 1982; Hawking, 1982, 1985; Starobinsky, 1982 ; Guth and
Pi, 1982 ; Bardeen et al., 1983). However, on a much greater scale
perturbations bp/p become very large. The estimates of bp(x)/p in the
chaotic inflation scenario show (Linde, 1984c) that density perturbations
formed at the moment at which the classical scalar field was equal to _ _ . MP ·
1 /2 (4.3)
m
Perturbations which are formed at that moment have at present the
wavelength
(4.4)
for m/Mp "' 10 - s . Eq . (4.4) gives the size of a locally-Friedmannian part of

our universe after inflation, which is much smaller than the classical estimate
l "' 101 0 10 cm, but is still many orders of magnitude greater than the size of
the observable part of the universe, "' 1028 cm. Let us try to understand the
origin of large inhomogeneities formed at ¢ -Z:, (M p/m) 1 1 2 · Mp , eq. (4.3), and
the global structure of the universe at scales greater than M; 1
exp [(2nMp/m)] , eq . (4.4).
Evolution of the fluctuating field <P in any given domain can be described
with the help of its distribution function P(<jJ), or in terms of its average value
<Ji in this domain and its dispersion ,'.\ = ( ( b¢2 ) ) 11 2. However, one will obtain
different results depending on the method of averaging : one can consider the
distribution Pc(¢) over the non-growing coordinate volume of the domain
(i.e. over its physical volume at some initial moment of inflation), or the
distribution PP(¢) over its physical (proper) volume, which grows
exponentially at a different rate in different parts of the domain. It can be
shown that the dispersion of the field ¢ in the coordinate volume i\c is always
much smaller than '<Jic for V((fic) � M� . Therefore the evolution of the
averaged field <Jic can be described approximately by eqs. (2.2)-(2.5).
However, if one wishes to know the resulting spacetime structure and the
distribution of the field ¢ after (or during) inflation, it is more appropriate to
take an average <JJP over the physical volume, and in some cases the
behaviour of <JJP and i\P differs considerably from the behaviour of <Pc and i\c .
To illustrate this let us note that perturbations b</J(x) generated during
inflation always have a wavelength i\l -z:, H - 1 (¢). Each domain of the
universe of a size .11 -z::, H - 1 (¢), which is the size of the event horizon in
de Sitter space, evolves independently of what occurs outside it (Gibbons
and Hawking, 1977; Hawking and Moss, 1982). Therefore, as a result of
generation of long-wave perturbations of the field ¢, during inflation the
universe becomes effectively divided into many mini-universes of initial size
D.l -Z::, H - 1 ( ¢ ). Each of these domains may be considered as a separate part of
a Friedmann universe with a coordinate volume which effectively does not
change during inflation. The field ¢ inside a domain of a size L\l"' H - 1 looks
as if it were almost exactly homogeneous [aµ (b</J) · aµ (b<jJ)] "' H4 � V(</J) for
V � M� and b</J given by eq. (4. 1 ). During a typical time L\t = H - 1 this field
6 14 A. Linde
decreases by
mMp M�
t1.</> (4. 5)
= 2J3n1 ' 2H = 4n</> '
in accordance with eq. (2.4). The physical size of this domain during the time
llt = H - 1 increases e times, and its physical volume increases e 3 times. As a
result of generation of perturbations of the field </>, eq. (4. 1), the value of the
field </> in this domain becomes </> - ll</> + b</>(x). Note that for
</> � Mp(Mp/m)1 1 2 , l b</>(x) I is much greater than ll</>, see Fig. 13. 1 . Since a
typical wavelength of the perturbations b</>(x) generated during the time
llt = H - 1 is O(H - 1), the domain whose initial size is O(H - 1), after
expanding e times, becomes divided into 0( e 3/2) domains of size O(H - 1 ),
m which the field has a value </> - O(H), and O(e3/2) domains of size
Fig. 1 3 . 1 . The region of possible values of the field Mi/m. Fluctuations of metric and of the scalar field <P are
seems impossible. (2) Mp(M p/m) 11 2 :5 <P :5 Mi/m. In this region fluctuations
extremely large, and any classical description of this field and of spacetime
of the field <P are large, but dispersion of these fluctuations averaged over the
coordinate volume is small, Ll � fP · I n this region fPc rolls down according to
c c
eq. (2.4), but the field <P averaged over the physical volume, fPp, diffuses up to
<P - Mi/m. (3) Mp :5 </J :5 Mp(Mp/m)1' 2 . In this region <Pc � <PP decreases
according to eq. (2.4). Inflation occurs until the field <P enters the region
<P � Mp. (4) <P � Mp. The field <P rolls down , oscillates near the minimum of
V(<P) and its energy transforms into heat.
v
Afp4
mM3p
0 Af 2
m
.:::.r
O(H - i ) with field value <P + O(H). In other words, the original domain
(mini-universe) of size O(H - 1 ) after the time dt = H - 1 , expands and
separates into O(e 3 ) mini-universes of a size O(H - 1 ), and in (almost)
half of these mini-universes the field <P grows rather than decreases. During
the next interval dt = H - 1 the total number of domains with a growing
field <P increases again, and so on. This means that the total physical
volume of domains containing permanently growing field <P increases as
exp[(3 - ln 2)Ht] - exp(2.3H(</J)t) for </J � Mp(Mp/m) 1 1 2, whereas the total
physical volume of domains in which the field <P does not decrease grows
approximately as ! exp(3Ht). Since the value of H(</J) increases with
a growth of ¢ , the main part of the physical volume of the universe
emerges as a result of expansion of domains with a maximal possible
field ¢ , i .e. with <P - M�/m, at which V(</J) - M�. (Note that for <P � M�/m,
if it is possible to consider such domains at a classical level, the process
of self-reproduction of inflationary mini-universes with a growing
field <P becomes suppressed, since at V(</J) � Mi a typical value of
aµ (b</J) aµ (b</J) - H 4 is greater than V(</J), which does not lead to creation of
inflationary mini-universes with <P � M�/m.) Therefore, whereas the field <P
averaged over the coordinate volume of any given domain (i.e. cpc),
gradually decreases in accordance with eq. (2.4), the field <P averaged over
the physical volume of a domain that initially contains the field
<P "<, Mp (Mp/m) 1 12 grows to cpp - M�/m (Linde, 1986a).
It may be useful to look at the same problem from another point of view.
The Brownian motion of the field <P at Mp � <P � M�/m can be. described (for
changes of the field <P that are not too large) by the. diffusion equation
=
at a¢
(
ape !_____ a(�Pc) � a v
a¢· + )
3H a¢ '
(4.6)
where the coefficient of diffusion � = H 3/8n2 . This equation for the case
H(</J) = const. was first derived by Starobinsky ( 1984, 1986); for a more
detailed derivation see Goncharov and Linde ( 1986, 1987). For the special
case a v;a¢ = 0 this equation was obtained by Vilenkin ( 1983).
The stationary solution (aPc/at = O) would be
Pc - exp(3M;/8 V(</J)), (4.7)
which is equal to the square of the Hartle-Hawking wave function of
the universe (3. 5) (Starobinsky, 1986; Linde, 1986b, c). A more general
stationary solution would also contain a term
3nj 0 �!) · exp (83�J) ) J: dq\ exp ( - �7/i) (4.8)
616 A. Linde
-
(Starobinsky, 1986), which would correspond to a stationary flux of
probability Jo from cf> = + oo to cf> = oo . However, eq. (4.6) actually has no
normalisable stationary solutions, since it is not valid at l ct>I � Mp (and at
l ct>I � M�/m).t The field cf> at l c/> I � Mp is rapidly oscillating rather than slowly
rolling down, there is no diffusion of the field cf> from the region cf> � Mp to
the region cf> � Mp , and the averaged field cf> decreases according to eq. (2.4).
A detailed discussion of the solutions of eq. (4.6) is included in a separate
publication (Goncharov and Linde, 1986, 1987 ; Linde, 1986c). Here we
would like to discuss some qualitative' features of these solutions and their
physical interpretation.
Let us consider a domain of initial size I - H - 1 (cf>). The field cf> = cf>o inside
such a domain is distributed homogeneously, since its classical part becomes
homogeneous due to inflation and perturbations b cf>(x) become essential
only at a scale I � H - 1 . There are two main stages of evolution of the field cf>
inside this domain. The first stage has a duration !lt ,...., 2 fo c/> 0/mM p .
During this time the average field (/Jc inside this domain remains practically
unchanged, (/)c = c/> 0 , eq. (2.4), whereas the dispersion 11-; = (bc/>2 ) grows as
(H 3/4n 2)!1t (Vilenkin and Ford, 1982; Linde, 1982c; Starobinsky, 1982) up to
the value 11; "" (cm2cf>ci/Mt), c = 0(1). Note that, at m2c/>�/2 4:. Mt, !le is much
smaller than ¢ 0 . Therefore the behaviour of the field (/Jc at the next stage of
the universe's expansion can be described by eq. (2.4), and the field (/Jc
decreases linearly. Fluctuations be/> are also generated at that stage, but they
have much smaller amplitude and dispersio!l, whereas the dispersion of
bcf>(x) produced at the first stage behaves as (/Jc (Guth and Pi, 1982), and it
therefore remains constant during inflation, l bcf>(x) I - l¢cl = mM p/2 fo =
const. As a result, the function Pc(cf>) is mainly determined by the
fluctuations produced at the first stage, and is given by
p (
c ( A-'+' ) ,...., ex p - A
2 Llc2
) (
(cf> - (/)c(t))2 (cf>
"' exp - - (/)c
)2Mt )
2Cm 2A..'f' 4O • (4.9)
Eq . (4 . 9) shows that the coordinate volume occupied by a large field cf> is

exponentially small, and it rapidly decreases for t > 2 fo cf> 0 /mMp (non
stationary regime!). On the other hand, during the time t ,...., 2 fo ¢ 0/
mMp , at which the distribution (4.9) remains unchanged, domains having a
large field cf> expand exp[B(¢ 2/M�)] times, B = O(l). This gives the
t Eq . (4.8) (without the term (4.7)) may represent such a solution, but only if there exists
a constant probability flux from the region </>'$> M�/m, where eq. (4.6) is not valid.
distribution Pp (</>) over the physical volume of the domain under

consideration after the time t ,..., 2fo </> 0/mM P :
p
P
(</>) -
(
exp -
2cm2</>6
)
(</> - </> 0 )2M� + 3B</>2
M� ·
(4. 10)
For </> 0 � Mp(M p/m ) 1 12 the distribution PP (</>) grows with increase of </>. This
means that after a time t - </> 0/mM P the field </> inside the main part of the
physical volume of a domain with ¢ 0 � M p (M p /m) 1 ' 2 grows and becomes
larger than </> 0 , in agreement with our previous results.
For completeness we shall mention here another solution of eq. (4.6). If
the initial value of the field </> is very large, ¢ 0 � M � /m, i.e. if one starts with
the spacetime foam with V(</> 0 ) � M� , then the evolution of the field </> in the
first stage (rapid diffusion) becomes more complicated (the naively
estimated dispersion 11; ,.,,, H 3 !lt soon becomes greater than </>5). In this case
the distribution of the field </> is not Gaussian. The solution of eq. (4.6) at the
( )
stage of diffusion from </> 0 to some field </> with V( </>) � M� is given by
Pc(</>) ,..., exp - fo
3 M:
. (4. 1 1)
m3</>t
This solution describes quantum creation of domains of a size l � H - 1 (</> ),
which occurs due to the diffusion of the field </> from </> 0 � M � /m to </> � ¢ 0 .
Direct diffusion with formation of a domain filled with the field </> is
possible only during the time t = c(2fo </J/�Mp ) , c = O ( l ) . At larger
times a more rapid process is a diffusion to some field $ > </> and a
subsequent classical rolling down from $ to </> . Therefore one may interpret
a distribution Pc (</J) formed after a time t = c(2fo </J/mM p ) as a probability
of a quantum creation of a mini-universe filled with a field </> (Linde,
1986b, c ; Goncharov and Linde, 1987)
(
Pc (</>) - exp - c
3M 4
2m2;2
) (
"' exp - 2c
3M 4
S V(;) ,
) (4. 12)
which is in agreement with the estimate (3 . 6) _ of the probability of a
quantum creation of the inflationary universe (Linde, 1984a, b; Zeldovich
and Starobinsky, 1984; Rubakov, 1984; Vilenkin, 1983). The same result is
valid for all sufficiently steep potentials V( </> ), in particular for all V (</> ) - </>"
(Linde, 1986b, c ; Goncharov and Linde, 1987), and, as discussed by
Rubakov ( 1984), no quantum particle production occurs during the process
of the mini-universe creation. Presumably, the process considered above is
complementary to the process of quantum creation 'from nothing' (see also
the next section). In our case, creation of an inflationary (mini-) unive�e
618 A . Linde
occurs as a diffusion ( = tunnelling) not from 'nothing' (whatever that

means), but from the spacetime foam with V(¢) � M� , which in our case fills
most of the physical volume of the universe and plays the role of the unstable
(but regenerating) gravitational vacuum.
13.5 Eternal chaotic inflation
From the results obtained in the previous section it follows that the
inflationary universe, which contains at least one domain of a size l �
H- 1 (¢), with ¢ �Mp(Mp/m) 11 2 , endlessly reproduces inflationary mini
universes with ¢ � Mp(Mp/m) 11 2 • The global geometry of the inflationary
universe has nothing in common either with the geometry of an open or flat
homogeneous universe with a gradually decreasing energy density or with
the geometry of a closed universe, which is created at some initial moment
t = O and which disappears as a whole at another moment t = tmax · t In our
case the universe endlessly regenerates itself, and there is no global 'end of
time'. Moreover, it is not necessary to assume that the universe as a whole
was created at some initial moment t = 0. The process of creation of each
new mini-universe with ¢ � Mp(Mp/m) 1 1 2 occurs independently of the pre
history of the universe, it depends only on the value of the scalar field inside a
domain of a siz� O(H - 1 (¢)) and not on the moment of creation of the mini
universe. Therefore the whole process can be considered as an infinite chain
reaction of creation and self-reproduction which has no end and which may
have no beginning.
A similar situation may occur in the old inflationary universe scenario
(Guth, 198 1 ; Guth and Weinberg , 1983) and in the new inflationary universe
scenario (Steinhardt, 1983; Linde, 1982e; Vilenkin, 1983). If the probability
of bubble production with <P =f. 0 is sufficiently small, or if the average time
for the field <P to roll down from the extremum of V(¢) at ¢ = 0 is sufficiently
large, then the phase transition to the global minimum of V(¢) occurs in the
major part of the coordinate volume of the universe, but the major part of
the physical volume of the universe remains in the unstable state ¢ = O for
ever. This may give us a solution to the problem of the initial cosmological
singularity : the major part of the inflationary universe may always exist in
the unstable but self-reproducing state ¢ = 0, and it is not necessary to
t This means in particular that the global geometry of the inflationary universe cannot
be studied in the context of the minisuperspace approach. However, the geometry of a
locally Friedmannian mini-universe and the process of its formation can be investigated
with the help of the minisuperspace approach, but in this approach formation of a mini
universe looks like creation of the universe 'from nothing'.
Itiflation and quantum cosmology 619
assume that the universe was created as a whole at some initial moment t =0
(Linde, 1982e).
There are some problems associated with this suggestion. The parts of the
universe with </> = 0 locally have the same geometrical properties as de Sitter
space. A geodesically complete de Sitter space is closed and its scale factor is
a(t) = H - 1 cosh Ht. For oo < t < 0, a(t) decreases, and the lifetime of such a
-
space in the unstable vacuum state </> = 0 is finite. Therefore such a universe
cannot remain in the unstable state </> = 0 at t = O, when inflation starts. One
may therefore wonder whether the eternally existing universe without initial
singularity and with the major part of its volume in the unstable state <f> = O,
is geodesically complete, or whether the initial singularity in this scenario is
unavoidable (Linde, 1983a).
One should note, that the global geometry of such a universe differs
considerably from the geometry of de Sitter space, since the part of the
coordinate volume which remains in the de Sitter phase <f> =O becomes
infinitesimally small in the course of time. Therefore it is not excluded that
such a universe is geodesically complete, which gives us a possible solution
of the problem of the initial cosmological singularity (Linde, 1982e).
However, there is no need to enter into a detailed discussion of this
possibility here, since no realistic versions of the old and of the new
inflationary universe scenario have been elaborated so far, and the
probability of quantum creation of the universe with </> = 0, V(O) ;S 10- 1 0 M�
(see the beginning of this chapter) is vanishingly small, see eq. (3.6).
On the other hand, in the eternally existing chaotic self-reproducing
inflationary universe discussed above the major part of the physical volume
of the universe emerges from the regions with V( </>) M � . In this sense the
-
universe is effectively singular. Thus we do not claim that the nonsingular

part of the universe in this scenario is geodesically complete. What is
important, however, is that these 'singularities' do not form a global space
like singular hypersurface, which would mean the existence of a 'beginning
of time' for the w hole universe. The existence of such a hypersurface does not
follow from general topological theorems concerning singularities in general
relativity (Hawking and Ellis, 1973), but it is usually assumed that our
universe looks like a slightly inhomogeneous Friedmann universe, in which
such a hypersurface does exist. However, as we have seen, the global
geometry of the inflationary universe has nothing in common with the
geometry of the Friedmann universe; a global space-like singular
hypersurface in the inflationary universe does not exist in the future and well
may not exist in the past.
620 A. Linde
Thus, there exist two main versions of the chaotic inflationary scenario .
(i) There may exist an initial global singular space-like hypersurface. In
this case the universe as a whole emerges from a state with a Planck density
p ,..,., M� at some moment t = tp , at which it becomes possible to speak about
the universe in terms of classical spacetime. A natural initial value of the field
¢ at the Planck time is ¢ "' M�/m "' 105Mp for m - 10- 5Mp. Then the
universe endlessly reproduces itself, due to generation of long-wave
fluctuations of the field ¢. This process occurs for M p(M p/m) 11 2 ;5 4> ;5 M�/m
(300Mp ;S c/> ;S M�/m) , i.e. for mM� ;S V(c/>) ;S M: . In the domains with ¢ ;5
Mp(Mp/m) 1 '2 this process becomes inefficient, and each such domain after
inflation looks like a Friedmann m1m-universe of a size / ,...,
Mi 1 exp [2n(Mp/m)] "' 10 1 0 5 cm. In this model the universe has a beginning
but has no end.
(ii) The possibility that the universe has a global singular space-like
hypersurface seems rather improbable unless the universe is compact and its
initial size is O(M; 1 ) . There is no reason for different, causally disconnected,
regions of the universe to start their expansion simultaneously. If the
universe is not compact, there should be no global beginning for its
evolution. A model which illustrates this possibility was suggested above.
The inflationary universe may infinitely reproduce itself, and it may have no
beginning and no end. Any two points of such a universe in a sufficiently
distant past could be causally connected and corresponding observers could
synchronise their clocks even though later they may live in mini-universes
which have become causally disconnected due to the exponential expansion
of the universe.
Note that the 'energy' of the scale factor a(t) is negative, and inflation may
be considered as a result of instability arising from the pumping of energy
from a(t) to the field c/>(t) (Linde, 1984c). For example, the energy of the
scalar field 4> in a closed inflationary universe grows exponentially, whereas
the energy of the scale factor a(t) becomes exponentially large and negative,
the sum of their energies being equal to zero. It was unclear why this
instability, being potentially possible, does not develop after inflation. We
now know a possible answer. The gravitational vacuum in the major part of
its physical volume always remains in the unstable inflationary state with
energy density of the order of M� . However, during the evolution of the
universe many islands of stability are formed, one of which is the mini
universe in which we now live.
Inflation and quantum cosmology 62 1
13.6 Global structure of the inflationary universe and the anthropic

principle
1 3 .6.1 The domain structure of the universe
The process of mini-universe formation and self-reproduction occurs at
densities mM� :S V(</J) :S M; , i.e. at 10 - sM; :S V(</>) :S M� for m "' 10 5 M p .
-
Note that this process may occur at densities many orders of magnitude
smaller than Mt . Therefore to prove the very existence of the process of self
reproduction of the inflationary universe in our scenario there is no need to
appeal to unknown physical processes at densities greater than the Planck
dens1ty P P "' M 4P .
.
On the other hand, it is very important that independently of the origin of

the universe in our scenario (whether the universe was created as a whole at
t "' tp , or whether it exists eternally), it now contains an exponentially large
(or even infinite) number of causally disconnected mini-universes, and a
considerable proportion of these mini-universes were created when the field
</> was O(M�/m) and its energy density was (almost) as large as M;. (Note
that this is true in the chaotic inflationary scenario only, in which inflation
may occur even at V(</>) "' M� .)
The local structure of spacetime inside a new-born mini-universe formed
at V(</>) � M; remains tinchanged. However, in realistic theories, in which
there exist many types of scalar fields with masses mill � H ( </>), large-scale
fluctuations of all these fields similar to the fluctuations of the field </> are
=
formed. Since the Hubble parameter H(</>) during mini-universe formation is
very large, H (8n V(</>)/3M�) 11 2 � 10 - 2 Mp "' 10 1 7 GeV, fluctuations of the
scalar fields are strong enough to transfer the classical fields in the new
born mini-universe from one local minimum of the effective potential
V(, </>) to another (Linde, 1983c, 1984c). This changes the low-energy
elementary particle physics inside the new mini-universe. A particular
example which is relevant to this effect is the supersymmetric S U(5) model.
The effective potential V() in this model has several minima of
approximately equal depth, and only one of them corresponds to the
desirable symmetry breaking S U(3) x U( 1). Even ifthe universe initially was
in one particular vacuum state corresponding to one of these minima of
V(), after inflation it becomes divided into many mini-universes
corresponding to all possible minima of V(). A typical time which is
necessary for a quantum tunneling from one local minimum of V() to
another is many orders greater than 10 1 0 years, which is the age of the
observable part of the universe. In one of these mini-universes the vacuum
622 A. Linde
state corresponds to the S U(3) x U( l) minimum of V(<l>). We live in this

mini-universe, not for the reason that the whole universe is in the S U(3) x
U( 1) symmetric state and not for the reason that the corresponding
minimum of V(<l>) is deeper than other minima, but for the reason that life of
our type is impossible in the mini-universes with other types of symmetry
breaking. This solves the problem of symmetry breaking in the
supersymmetric S U(5) model (Linde, 1983c).
In order to illustrate new possibilities which appear in the context of the
scenario discussed above, let us consider now a toy model which may
explain the small value of the vacuum energy density Pv (of the cosmological
constant), in the observable part of the universe ( !Pv I � 10 - 29 g cm - 3 ). This
model is probably unrealistic, but nevertheless it may be rather instructive.
The model describes an inflaton field ¢, which drives inflation, and a field <l>
with an extremely flat effective potential, V(<l>) = aM�(<l> - C), where a �
10 - 1 2 0 . It can be shown that during the time interval t - 10 1 0 years after
inflation such a field <l> remains essentially unchanged due to the very small
slope of V(<l>). However, the Brownian motion of this field during inflation is
very rapid, and it divides the universe into an exponentially large number of
mini-universes containing all possible values of the field <l> for which I V(¢) +
V(<l>)j � M: . After inflation the vacuum energy Pv inside these mini-universes
is given by V(<l>, ¢) = aM�(<l> - C) + V(¢ 0 ), where ¢ 0 corresponds to the
minimum of V(¢ ). This quantity in different mini-universes changes
·continuously from - M� to + M: , but life of our type is possible only in
those mini-universes in which I Pv I � 10 - 29 g cm - 3 . Indeed, domains with
V(<l>, ¢) < 0, j V(<l>, ¢)j � l0 - 29 g cm - 3 correspond to anti-de Sitter mini
universes with a lifetime -r � 10 1 0 years. The domains with V(<l>, ¢) �
10 - 29 g cm - 3 remain inflationary for t < 10 1 0 years, and the present density
of matter in such domains is negligibly small. For this reason we see
ourselves inside a domain with IPv l � 10 - 29 g cm - 3 .
To make this model realistic it would be desirable to explain why V(<l>) is
so flat (though similar potentials sometimes appear in realistic models
describing unified theories of elementary particles). In any case, the main
idea suggested above m�.y� b�used in other models as well . In our scenario it
is not necessary to insist (as is usually done) that the vacuum energy must
disappear in a 'true' vacuum state. It is quite sufficient if there exists some
relatively stable vacuum-like state with 1 Pv l � 10 - 29 g cm- 3 . This
requirement is still very restrictive, but it can be satisfied much more easily
than the previous one. For a discussion of a similar approach to the
cosmological constant problem, see also Sakharov ( 1984) and Linde

(1984c) .
1 3 .6.2 A scenario of inflationary compactification in the Kaluza-

K lein cosmology
In the mini-universes of initial size O(H - t ) � O(M; 1 ), which are formed at
V( ¢ ) - M� , fluctuations of the metric on a scale O(M; 1 ) are of the order of
unity. Therefore, if a theory of the type of eq. (2.1) can be considered as a part of
a Kaluza-K.lein (or superstring) theory, then at the moment of mini-universe
formation the type of compactification and the number of compactified
dimensions inside the new-born mini-universe can be changed (almost)
independently of what occurs in causally disconnected regions outside it.
(The only possibly constraint on the local changes of spacetime structure
stems from topological considerations and may lead, for example, to the
formation of a pair of mini-universes with opposite topological numbers.)
As a result, our universe becomes divided into an exponentially large
number of domains, and all possible types of compactification and all
possible (metastable) vacuum states should exist in different domains of the
universe. All these domains (mini-universes), in which inflation remains
possible after their formation, later become exponentially large (Linde,
1986a, b, c).
For a detailed investigation of this possibility it would be desirable to
have a realistic model of inflation in Kaluza-Klein or superstring theories.
Unfortunately, the models of Kaluza-Klein inflation that now exist are not
completely satisfactory (Maeda and Pollock, 1986; Linde, 1986c). On the
other hand, we do not know any alternative possibility for solving all the
problems that are solved by the inflationary universe scenario. Therefore
one may consider the difficulties with the Kaluza-Klein (and superstring)
inflation as difficulties of these theories (Maeda and Pollock, 1986).
Actually, however, our understanding of these theories is incomplete, and
hopefully the corresponding problem will be resolved in future. At present
we are just trying to understand some possible features of the Kaluza-Klein
inflationary cosmology, which are rather unusual. To illustrate some of the
ideas discussed above, let us again consider inflation in the toy model, eq.
(2. 1).
In our investigation of inflation in this model we have studied only four
dimensional de Sitter-like solutions dS4 with metric
ds2 � dt2 - H - 2 cosh2 Ht(dx1 + sin2 x(dll2 + sin2 () d¢1)), (6. 1)
624 A. Linde
where H2 = 8 n V(</>)/3M� . However, there exist some other interesting
=
solutions, such as
ds 2 dt2 -1H - 2((cosh2(J3 Ht) dx2 + d82 + sin 2 8 d</>2). (6.2)
This solution describes a universe that is a product of a two-dimensional
de Sitter space dS and a compact sphere S with a very small radius R 2
2 2
(j3 H) - 1 . This is a particular case of a Kantowski-Sachs universe
=
(Kantowski and Sachs, 1966; Kofman et al. , 1983; Paul et al., 1986). This
solution is unstable; it (locally) transforms into dS4. However, it is possible
to stabilize it, for example by an appropriate addition of Rn terms to the
Lagrangian, eq. (2. 1) (Deruelle and Madore, 1986 ; Linde and Zelnikov,
1987).
Recently it was shown that long-wave quantum perturbations of scalar
fields </> similar to those discussed in Section 13.4 are generated in the
dS 2 x S 2 universe (6.2) as well (Kofman and Starobinsky, 1987). t This result
is directly related to the main conclusions of the present chapter (Linde,
1986c). Namely, if inflation is initially three dimensional as in ordinary de
Sitter space dS 4 (6. 1), then fluctuatio'n s of the scalar field </> create some
inflationary domains (mini-universe), in which V(</>) ,..., M;. Due to large
fluctuations of </> and of metric gµ11 inside domains of a size O(H - 1 ) ,..., Mi 1 ,
expansion inside some of these domains may become one dimensional (6.2),
independently of what occurs in the nearby domains. This leads to formation
of Kantowski-Sachs 'branches', which spread out of de Sitter space.
However, fluctuations in the Kantowski-Sachs mini-universes lead again to
creation of domains with V(</>) ,..., M; where de Sitter mini-universes (6. 1) can
be created. With account taken of this effect, the inflationary universe in our
scenario may appear as a system of huge inflationary bubbles dS4 connected
with each other by thin inflationary tubes dS 2 x S 2 of exponentially large
length (Fig. 13.2).
The possibility of existence of such a complicated spacetime structure is
directly related to the 'no-hair' theorem for de Sitter space, which is valid
also for the one-dimensional exponential expansion (6.2). The processes
which occur in a part of a tube (6.2) of initial size dl <, H - 1 proceed
independently of what occurs in other parts of the universe. Actually, the
Kantowski-Sachs universe (6.2) is not the only solution of the Einstein
equations with V(</>) > 0 which leads to spontaneous compactification of a
t This is actually a general property of all inflationary models related to the existence of
d zero modes of a scalar field </> in the Euclidean section Sd of d-dimensional de Sitter
space dSd .
four-dimensional space during inflation. We hope to return to the discussion

of this question in a separate publication. Here we would only like to
mention that the effect considered above is fairly general. For example, in d
dimensional space (d > 4) inflation and fluctuations of scalar fields lead to
formation of a complicated structure consisting of inflationary mini
universes of all possible types, including mini-universes dSd, dSd - 2 x S 2 ,
dSd - J x S 3 , etc., connected with each other.
During inflation the radius R11 of each compactified sphere S11 remains of
Fig . 1 3 .2. This picture gives some idea of the global structure of the chaotic
inflationary universe. Child ; universes created where V(<J>) � M; have the
same 'genetic code' as their mother universes : they have the same number of
dimensions and the same (or almost the same) vacuum structure. However,
the universes created where V(<J>) is not much smaller than M; are
'mutants', which may have different dimensionality and different low-energy
elementary particle physics inside them. Each mini-universe may die, but the
universe as a whole has no end and may have no beginning. A typical
thickness of 'tubes' connecting mini-universes after inflation may become
very large . However, during inflation some of the tubes have a very small
thickness. For example, some of the de Sitter mini-universes dS4 may be
connected by the Kantowski-Sachs tubes dSd - n x Sn, where the radius of the
sphere Sn is of the order O(H - 1 (</>)). The mini-universes compactified during
inflation may ·serve as seeds for the next stage of the process of
compactification, which occurs after inflation.
626 A . Linde
the order H - 1 ( ¢ ), which initially is as small as Mi 1 . The main part of the

physical volume of the universes dSd - n x S11 remains in the state with V(¢) "'
M: for ever due to the effects connected with the long-wave fluctuations of
the field ¢, see Section 13.4. In those domains of the universe in which th e
field ¢ slowly decreases, the radius of compactification R11(¢) slowly grows
as H - 1 ( ¢) unless there exist some dynamical mechanism which fixes the
radius of compactification at some value R 0 � M; 1 . In this sense the
domains of the universe compactified during inflation may serve as seeds for
different possible types of compactified spaces in the Kaluza-Klein theories.
This provides us with a new scenario of compactification as compared
with the scenario of a classical power-law Kasner-like anisotropic expansion
discussed by several authors (see, e.g . Freund, 1982; Freund and Oh, 1985).
Namely, during inflation at mM� � V(c/J) � M: mini-universes with all
possible types of compactification are formed in different causally
disconnected regions of the universe, and those mini-universes in which
inflation (in the uncompactified directions) remains possible after their
formation later become exponentially large. The process of formation of
new mini-universes has no end, in the major part of the physical volume of
the universe _this process occurs even now, and therefore even if the
probability of formation of a mini-universe of some particular type is
strongly suppressed, many such mini-universes should exist at present.
According to this scenario, we live in a four-dimensional spacetime with
our type of compactification , not for the reason that other types of
compactification are impossible (or improbable), but for the reason that life
of our type cannot exist in spaces with other dimensionality and with a
different low-energy elementary particle physics (Linde, 1983a, 1984c,
1986c; see also Rozental, 1980, 1984).
13 .6 .3 Inflation and the anthropie principle
Discussion of some of the problems considered above is based on the so
called anthropic principle (Dicke, 196 1 ; Collins and Hawking, 1973 ; Carr
and Rees, 1979; Rozental, 1980; Linde, 1984c). According to this principle,
we live in a four-dimensi onal, homogeneou s, isotropic world with
e2/4n = 1 �7, me = 0.5 MeV , etc. , for the reason that life of our type in a
different world would be impossible. For example, life of our type would be
our spacetime had dimensionality different from d 4. =

impossible if e, me, M p , etc., were half (or less) of their present values, or if
Several years ago the anthropic principle seemed rather esoteric, since it
implied that many universes might exist, but we live in just one of them, which
Injlation and quantum cosmology 627
is sufficiently suitable for us. It was not clear in what sense one could speak
of many universes if our universe is unique, and whether such fundamental
constants of nature as the dimensionality of spacetime, the vacuum energy,
the value of electric charge, etc. , can change when one 'travels' from one
universe to another.
A possible answer to the first part of this question was suggested by the
many-worlds interpretation of quantum mechanics (Everett, 1957). Further
investigation of this possibility may be of profound importance. However, it
seems doubtful that it is possible to use this interpretation correctly without
a proper understanding of the nature of consciousness. Does an observer
just observe the universe, or does he 'create' it? (Wheeler, 1979). What is
actually split : the universe or consciousness? Can consciousness exist 'by
itself' Oike spacetime without matter) or is it merely an arena for the
manifestation of spacetime and matter? In what sense can consciousness
'choose' a universe to live in?
Thus the development of physics reveals problems which traditionally
were beyond the scope of physics . It seems that to go further we must
investigate these problems without prejudice, rather than wait until
philosophers try to do it for us. However, such a path is not easy to follow,
and it seems encouraging that inflation makes it possible to circumvent
some of the above-mentioned problems that precluded the justification of
the anthropic principle. Namely, from the scenario discussed in this chapter
it follows that even if the universe at some time contains only one domain
(mini-universe) of initial size O(H<;O � O(Mp- 1 ), later it splits into many
causally disconnected mini-universes of exponentially large size, in which all
possible types of compactification and all possible vacuum states are
realised . In this sense our universe actually consists of many universes of all
possible types . Whereas several years ago the dimensionality of spacetime,
the vacuum energy density, the value of electric charge, the Yukawa
couplings, etc. , were regarded as true constants, it now becomes clear that
these 'constants' actually depend on the type of compactification and on the
mechanism of symmetry breaking, which may be different in different
domains of our universe .
One of the main objections to the anthropic principle was the assertion
that for the existence of life of our type there is no need for our universe to be
so uniform on scales much greater than the scale of the Solar System. We
now know the answer to this question . The size of the homogeneous locally
Friedmannian mini-universes in which bp/p ,....., 10 - 4 (which is necessary for
the formation of galaxies of our type) becomes after inflation much greater
628 A. Linde
than the size of the Solar System. For example, in the model (2. 1) bp/p on a
galactic scale is given by O( lO)(m/Mp), which gives m - 10 - 5Mp . On the
other hand, the size of a uniform part of our universe is greater than
Mi 1 exp [2n{Mp/m)] � 101 05 cm, which is much greater than the observable
part of the universe ( - 1028 cm).
This means that actually it is possible to justify some kind of weak
anthropic principle in t he inflationary cosmology. The line of thought
advocated here is an alternative to the old assumption that in a 'true' theory
it must be possible to compute unambiguously all masses, coupling
constants, etc. From our point of view, it is rather improbable that a 'true'
theory must have only one 'true' ground state. In the context of the unified
theories of all fundamental interactions, the validity of this assumption
becomes unlikely. According to the scenario discussed in this chapter, this
assumption is probably incorrect; in any case it is not necessary. The old
question of why our universe is the only possible one now is replaced by a
new question : in which theories is the existence of mini-universes of our type
possible? This question is still very difficult, but it is much easier than the old
one.
13. 7 Conclusions
The inflationary universe scenario continues to develop rapidly. It appears
that the global structure of the universe according to this scenario is
determined by quantum effects, that the major part of the physical volume of
the universe should even now be in an inflationary state with V(¢) - M� ,
that the universe eternally reproduces itself, and that it consists of
exponentially many mini-universes of different types. Some of these
results are model-dependent, some are not. We do not know what the fate of
the ideas discussed above will be, nor how they will be modified by future
developments of elementary particle physics and of the theory of
superstrings. In any case, the inflationary universe scenario may serve as a
good example, showing how many exciting surprises the theory of gravity
developed three hundred years ago can still give us.
References
Albrecht, A. and Steinhardt, P. J . ( 1982) . Phys. Rev. Lett., 48 , 1220.

Bardeen, J . , Steinhardt, P. J . and Turner, M. ( 1983). Phys. Rev., D28, 679.
Carr, B . J. and Rees, M . J. ( 1979) . Nature, 278, 605 .

Collins, C. B. and Hawking, S. W . ( 1973). Astrophys. J., 180, 3 17.
Deruelle, N. and Madore, T . ( 1986). Mod. Phys. Lett., Al , 237.
DeWitt , B . S . ( 1967). Phys. Rev., 160, 1 1 13 .
Dicke, R. H . ( 196 1). Nature, 1 92 , 440.
Dowker, J . S . and Critchley, R. ( 1976). Phys. Rev., 0 13, 3224.
Dreitlein, J . ( 1 974) . Phys. Rev. Lett. , 33, 1243 .
Enquist, K . , Nanopoulos, D. V . , Papantonopoulos, E. and Tamvakis, K. ( 1986). Phys.
Lett., 166B, 4 1 .
Everett, H . ( 1957). Rev. Mod. Phys., 29, 454.
Freund , P. G. 0. ( 1982). Nucl. Phys., B209, 146 .
Freund, P . G . 0 . and Oh, P . ( 1985). Nucl. Phys., B255, 688.
Gibbons , G. W . and Hawking , S . W . ( 1977) . Phys. Rev., 015, 2738.
Gliner, E. B. ( 1965). Sov. Phys.-JETP, 22, 378 .
Gliner, E. B. ( 1970). Dok/. Akad. Nauk SSSR , 192, 77 1 .
Gliner, E . B. and Dymnicova , V. ( 1975). Pis'ma V. Astron. Zh . , 1 , 5, 7 .
Goncharov, A . S . and Linde, A . D. ( 1986). Fiz. Elem . Chast. Atom Yad., 17, 837. Sov. J .
Particles and Nuclei, to b e translated .
Goncharov, A . S . and Linde, A . D. ( 1987) . ZhETF, 92 , 1 137.
Gurevich, L. E. ( 1975). Astrophys. Space Sci., 38, 67.
Guth, A . H . ( 198 1). Phys. Rev., 023, 347.
Guth, A . H . and Pi , S .-Y. ( 1982) . Phys. Rev. Lett. , 49, 1 1 10.
Guth, A. H . and Weinberg , E. ( 1983). Nucl. Phys., B21 2 , 32 1 .
Hartle, J . B. and Hawking, S . W . ( 1983) . Phys. Rev., 028 , 2960.
Hawking, S. W. ( 1982). Phys. Lett., 1 15B, 295.
Hawking, S . W . ( 1984) . Nucl. Phys., B239, 257.
Hawking, S . W . ( 1985). Phys. Lett. , 150B, 339.
Hawking, S. W . and Ellis, G . F. R. ( 1973). The Large Scale St1;ucture of the Universe.
Hawking, S. W. and Moss, I. G . ( 1982) . Phys. Lett., HOB, 35.
Hawking, S. W . and Page, D . N . ( 1986) . Nucl. Phys., B264 , 185.
Kantowski , R . and Sachs, R . K . ( 1966). J. Math. Phys., 7, 443 .
Kirzhnits, D. A . ( 1972) . JETP Lett., 1 5 , 529.
Kofman , L . A . , Sahni, V. and Starobinsky, A. A. ( 1983). ZhETF, 85, 1876.
Kofman , L . A . and Starobinsky , A . A . ( 1987). Submitted to Phys. Lett.
Linde, A. D. ( 1974). JETP Lett., 19, 183.
Linde, A. D. ( 1979). Rep. Prog . Phys., 42, 389.
Linde, A. D. ( 1982a). Phys. Lett., 108B, 389.
Linde, A . D. ( 1982b). Phys. Lett. , 1 14B, 43 1 .
Linde, A . D . ( 1982c). Phys. Lett., 1 16B, 335.
Linde, A. D. ( 1982d). Phys. Lett., 1 16B, 340.
Linde, A. D. ( 1982e). Non-singular regenerating inflationary universe. Cambridge
University preprint.
Linde, A . D. ( 1983a). I n The Very Early Universe, ed. G. W. Gibbons, S . W. Hawking
and S . Siklos, p. 205 . Cambridge University Press : Cambridge.
Linde, A. D. ( 1983b). Phys. Lett., 1 23B, 177.
Linde, A. D. ( 1983c) . Phys. Lett. , 131B, 330.
Linde, A. D. ( 1984a). Sov. Phys.-JETP, 60, 2 1 1 .
Linde, A. D. ( 1984b). Lett. Nuovo Cimento, 39, 40 1.
Linde, A. D. ( 1984c). Rep. Prog. Phys., 47, 925.
630 A. Linde
Linde, A. D. ( 1 985a). Phys. Lett., 1628, 28 1 .

Linde, A . D . ( 1985b). Prog. Theor. Phys. Suppl., 85, 279.
Linde, A. D. ( 1 986a). Mod. Phys. Lett., A l , 8 1 .
Linde, A . D . ( 1986b). Phys. Lett., 1 758, 395.
Linde, A. D . ( 1986c). Eternally existing self-reproducing inflationary universe. Trieste
preprint IC/86/75, to be published in the Proceedings of the Nobel Symposium on
Unification of Fundamental Interactions, Physica Scripta, 34.
Linde, A. D. and Zelnikov, M . (1 987). To be published.
Maeda, K. and Pollock , M. D. ( 1 986). Phys. Lett., 1738, 25 1 .
Mukhanov, V . F . and Chibisov, G . V . ( 1 98 1). JETP Lett. , 33, 532.
Mukhanov, V . F. and Chibisov, G. V. (1982). Sov. Phys.-JETP, 56, 258.
Paul , B . G., Datta, D. P. and Mikherjee, S. ( 1986). Mod. Phys. Lett., A l , 149.
Rozental, I. L. ( 1980). Sov. Phys. Uspekhi, 23, 296.
Rozental, I . L . ( 1984). Elementary Particles and Structure of the Universe. Nauka :
Moscow. To be translated into English by Springer : Heidelberg.
Rubakov, V . A. ( 1984) . Phys. Lett., 1488, 280.
Sakharov, A. D. ( 1984). Sov. Phys.-JETP, 60, 2 14.
Shafi, Q . and Wetterich, C. ( 1985). 1528, 5 1 .
Starobinsky, A . A. ( 1 979). JETP Lett., 30, 682.
Starobinsky, A. A. ( 1980). Phys. Lett. , 918, 99.
Starobinsky, A. A. ( 1 982). Phys. Lett., 1 178, 1 7 5 .
Starobinsky, A . A. ( 1 984) . In Fundamental Interactions, p. 55. M GPI Press : Moscow.
Starobinsky, A. A. ( 1 986). In Current Topics in Field Theory, Quantum Gravity and
Strings, Leet. Notes in Physics, ed. H. J. de Vega and N. Sanchez, 246, 107 . Springer:
Heidelberg.
Steinhardt, P. J. ( 1983). In The Very Early Universe, ed. G. W. Gibbons, S. W. Hawking
and S. Siklos, p. 25 1 . Cambridge University Press : Cambridge.
Turner, M . S. ( 1985). In Proceedings of the Cargese School on Fundamental Physics and
Cosnwlogy, ed. J. Audouze and J. Tran Thanh Van. Editions Frontiers : Gif-Sur
Yvette.
Veltman, M. ( 1974). Rockefeller University preprint.
Veltman, M . ( 1 975). Phys. Rev. Lett. , 34, 77.
Vilenkin, A. ( 1983). Phys. Rev., D27, 2848 .
Vilenkin, A. and Ford , L . H . ( 1982) . Phys. Rev., D26, 123 1 .
Wheeler, J. A. ( 1968). I n Batte/le Recontres, ed. C . DeWitt and J . A . Wheeler. Benjamin :
New York.
Wheeler, J. A. ( 1979). In Astrofizica e Cosmologia Gravitazione Q uanti e Relativita.
Guinti Barbera : Firenze.
Zeldovich, Ya. B. and Starobinsky, A. A. ( 1984). Sov. Astron . Lett., 10, 135.
14
Q uantum cosmology
S . W. H A W K I N G
14. 1 Introduction
A few years ago I received a reprint request from an Institute of Quantum
Oceanography somewhere in the Soviet Far East. I thought : What could be
more ridiculous? Oceanography is a subject that is pre-eminently classical
becau se it describes the behaviour of very large systems. Moreover,
oceanography is based on the Navier-Stokes equation , which is a classical
effective theory describing how large numbers of particles interact according
to a more basic theory, quantum electrodynamics. Presumably, any
quantum effects would have to be calculated in the underlying theory.
Why is quantum cosmology any less ridiculous than quantum
oceanography? After all , the universe is an even bigger and more classical
system than the oceans. Further, general relativity, which we use to describe
the universe, may be only a low energy effective theory which approximates
some more basic theory, such as string theory.
The answer to the first objection is that the spacetime structure of the
universe is certainly classical today, to a very good approximation.
However, there are problems with a large or infinite universe, as Newton
realised . One would expect the gravitational attraction between all the
different bodies in the universe to cause them to accelerate towards each
other. Newton argued that this would indeed happen in a large but finite
universe. However, he claimed that in an infinite universe the bodies would
not all come together because there would not be a central point for them to
fall to . This is a fallacious argument because in an infinite universe any point
can be regarded as the centre. A correct treatment shows that an infinite
universe can not remain in a stationary state if gravity is attractive. Yet so
firmly held was the belief in an unchanging universe that when Einstein first
proposed general relativity he added a cosmological constant in order to
632 S. W Hawking
obtain a static solution for the universe, thus missing a golden opportunity
to predict that the universe should be expanding or contracting. I shall
discuss later why it should be that we observe it to be expanding and not
contracting.
If one traces the expansion back in time, one finds that all the galaxies
would have been on top of each other about 15 thousand million years ago.
At first it was thought that there was an earlier contracting phase and that
the particles in the universe would come very close to each other but would
miss each other. The universe would reach a high but finite density and
would then re-expand (Lifshitz and Khalatnikov, 1963). However, a series of
theorems (Hawking and Penrose, 1970; Hawking and Ellis, 1973) showed
that if classical general relativity were correct, there would inevitably be a
singularity at which all physical laws would break down. Thus classical
cosmology predicts its own downfall. In order to determine how the classical
evolution of the universe began one has to appeal to quantum cosmology
and study the early quantum era.
But what about the second objection? Is general relativity the
fundamental underlying theory of gravity or is it just a low energy
approximation to some more basic theory? The fact that pure general
relativity is not finite at two loops (Goroff and Sagnotti, 1985) suggests it is
not the ultimate theory. It is an open question whether su pergravity, the
supersymmetric extention of general relativity, is finite at three loops and
beyond but no-one is prepared to do the calculation. Recently, however,
people have begun to consider seriously the possibility that general relativity
may be just a low energy approximation to some theory such as
superstrings, although the evidence that superstrings are finite is not, at the
moment, any better than that for supergravity.
Even if general relativity is only a low energy effective theory it may yet be
sufficient to answer the key question in cosmology : Why did the classical
evolution phase of the universe start off the way it did? An indication that
this is indeed the case is provided by the fact that many of the features of the
universe that we observe can be explained by supposing that there was a
phase of exponential 'inflationary' expansion in the early universe. This is
described in more detail in the articles by Linde, and Blau and Guth
(Chapters 13, 12, this volume). In order not to generate fluctuations in the
microwave background bigger than the observational upper limit of 10- 4,
the energy density in the inflationary era cannot have been greater than
about 10- 1 0 m� (Rubakov et al., 1982 ; Hawking, 1984a). This would put the
inflationary era well inside the regime in which general relativity should be a
Quantum cosmology 633
good approximation. It would also be well inside the region in which any
possible extra dimensions were compactified. Thus it might be reasonable to
hope that the saddle point or semi-classical approximation to the quantum
mechanical path integral for general relativity in four dimensions would give
a reasonable indication of how the universe began. In what follows I shall
assume that the lowest-order term in the action for a spacetime metric is the
Einstein one, as it must be for agreement with ordinary, low energy,
observations. However, I shall bear in mind the possibilities of higher-order
terms and extra dimensions.
14.2 The quantum state of the universe
I shall use the Euclidean path integral approach. The basic assumption of
this is that the 'probability' in some sense of a positive definite spacetime
metric g µv and matter fields <P on a manifold M is proportional to exp( - [)
where [ is the Euclidean action. In general relativity
f = __l_ J, R (g)1 12 ddx + -l Jal K(h) 1 12 dd - 1x - fLm [<P] (g)112 ddx ,
16n
M Sn
M
where h and K are respectively the determinant of the first fundamental form
and the trace of the second fundamental form of the boundary aM of M. In
string theory the action [of a metric 9µ v • antisymmetric tensor field Bµ v and
dilaton field <P is given by the log of the path integral of the string action
over all maps of string world sheets into the given space. For most fields the
path integral will not be conformally invariant . This will mean that the path
integral diverges and [ will be infinite. Such fields will be suppressed by an
infinite factor. However, the path integral over maps into certain
background fields will be conformally invariant. The action for these fields
will be that of general relativity plus higher-order terms.
The probability of an observable 0 having the value A can be found by
summing the projection operator n A over the basic probability over all
Euclidean metrics and fields belonging to some class C.
P0 (A) = L d[gµ,] d [</>] IT A exp( - f),
where n A = 1 if the value of 0 is A and zero otherwise. From such
probabilities and the conditional probability, the probability of A given B,
P(A, B)
P(A I B) =
P(B) ,
where P(A, B) is the joint probability of A and B, one can calculate the
outcome of all allowable measurements.
634 S. W. Haw king
The choice of the class C of metrics and fields on .w hich one considers the
probability measure exp( - l) determines the quantum state of the universe.
C is usually specified by the asymptotic behaviour of the metric and matter
fields, just as the state of the universe in classical general relativity can be
specified by the asymptotic behaviour of these fields. For instance, one could
demand that C consist of all metrics that approach the metric of Euclidean
flat space outside some compact region and all matter fields that go to zero
at infinity. The quantum state so defined is the vacuum state used in S matrix
calculations. In these one considers incoming and outgoing states that differ
from Euclidean flat space and zero matter fields at infinity in certain ways.
The path integral over all such fields gives the amplitude to go from the
initial to the final state.
In these S matrix calculations one considers only measurements at infinity
and does not ask questions about what happens in the middle of the
spacetime. However, this is not much help for cosmology : it is unlikely that
the universe is asymptotically flat, and, even if it were, we are not really
interested in what happens at infinity but in events in some finite region
surrounding us. Suppose we took the class C of metrics and matter fields
that defines the quantum state of the universe to be the class described above
of asymptotically Euclidean metrics and fields. Then the path integral to
calculate the probability of a value of an observable 0 would receive
contributions from two kinds of metrics. There would be connected
asymptotically Euclidean metrics and there would be a disconnected metric
which consisted of a compact component that contained the observable 0
and a separate asymptotically Euclidean component. One can not exclude
disconnected metrics from the class C because any disconnected metric can
be approximated arbitrarily closely by a connected metric in which the
different components are joined by thin tubes with negligible action. It turns
out that for observables that depend only on a compact region the dominant
contribution to the path integral comes from the compact regions of
disconnected metrics. Thus, as far as cosmology is concerned, the
probabilities of observables would be almost the same if one took the class C
to consist of compact metrics and matter fields that are regular on them.
In fact, this seems a much more natural choice for the class C that defines
the quantum state of the universe. It does not refer to any unobserved
asymptotic region and it does not involve any boundary or edge to
spacetime at infinity or a singularity where one would have to appeal to
some outside agency to set the boundary conditions. It would mean that
spacetime would be completely self contained and would be determined
completely by the laws of physics : there would not be any points where the
laws broke down and there would not be any edge of spacetime at which
unpredictable influences could enter the universe. This choice of boundary
conditions for the class C can be paraphrased as : 'The boundary condition
of the universe is that it has no boundary' (Hawking, 1982; Hartle and
Hawking, 1983 ; Hawking, 1984b).
This choice of the quantum state of the universe is very analogous to the
vacuum state in string theory which is defined by all maps of closed string
world sheets without boundary into Euclidean flat space. More generally,
one can define a 'ground' state of no string excitations about any set of
background fields that satisfy certain conditions by all maps of closed string
world sheets into the background. Thus one can regard the 'no boundary'
quantum state for the universe as a 'ground' state (Hartle and Hawking,
1983). It is, however, different from other ground states. In other quantum
theories non-trivial field configurations have positive energy. They therefore
cannot appear in the zero energy ground state except as quantum
fluctuations. In the case of gravity it is also true that any asymptotically flat
metric has positive energy, except flat space, which has zero energy.
However, in a closed, non-asymptotically flat universe there is no infinity at
which to define the energy of the field configuration. In a sense the total
energy of a closed universe is zero : the positive energy of the matter fields
and gravitational waves is exactly balanced by the negative potential energy
which arises because gravity is attractive. It is this negative potential energy
that allows non-trivial gravitational fields to appear in the 'ground' state of
the universe.
Unfortunately, this negative energy also causes the Euclidean action !for
general relativity to be unbounded below (Gibbons et al., 1978), thus
causing exp( - l) not to be a good probability measure on the space C offield
configurations. In certain cases it may be possible to deal with this difficulty
by rotating the contour of integration of the conformal factor in the path
integral from real values to be parallel to the imaginary axis. However, there
does not seem to be a general prescription that will guarantee that the path
integral converges. This difficulty might be overcome in string theory where
the string action is positive in Euclidean backgrounds. It may be, however,
that the difficulty in making the path integral converge is fundamental to the
fact that the 'ground' state of the universe seems to be highly non-trivial. In
any event it would seem reasonable to expect that the main contribution to
the path integral would come from fields that are near stationary points of
the action l, that is, near solutions of the field equations.
636 S. W. Hawking
It should be emphasised that the 'no boundary' condition on the metrics

in the class C that defines the quantum state of the universe is just a
proposal : it cannot be proved from something else. It is quite possible that
the universe is in some different quantum state though it would be difficult to
think of one that was defined in a natural manner. The 'no boundary'
proposal does have the great advantage, however, that it provides a definite
basis on which to calculate the probabilities of observable quantities and
compare them with what we see. This basis seems to be lacking in many
other approaches to quantum cosmology in which the assumptions on the
quantum state of the universe are not clearly stated . For instance, Vilenkin
( 1986) defines the quantum state in a toy minisuperspace model by requiring
that a certain current on minisuperspace be ingoing at one point of the
boundary of minisuperspace (corresponding to 'creation from nothing') and
outgoing elsewhere on the boundary (annihilation into nothing?) . However,
he does not seem to have a general prescription that would define the
quantum state except in simple minisuperspace cases. Moreover, his state is
not CPT invariant, which is a property that one might think the quantum
state of the universe should have. Similarly, Linde ( 1985 ; Chapter 13, this
volume) does not give a definition of the quantum state of the universe. He
also suggests that the Wick rotation for the Euclidean action of the
gravitational field should be in the opposite direction to that for other fields.
This would be equivalent to changing the sign of the gravitational constant
and making gravity repulsive instead of attractive.
14.3 The density matrix

One thinks of a quantum system as being described by its state at one time .
In the case of cosmology, 'at one time' can be interpreted as on a spacelike
surface S. One can therefore ask for the probability that the metric and
matter fields have given values on a d 1 su rface S. In fact , it is meaningful to
-
ask questions only about the d - 1 metric hii induced on S by the d metric 9 µ v
on M because the components nµgµv of 9 µ v that lie out of S can be given any
values by a diffeomorphism of M that leaves S fixed . Thus the probability
that the surface S has the induced metric hii and matter fields ¢0 is
P(h ,; , </> 0 ) = L [g,,]

d d [ </> 0] exp( l) n 1h,,. ;.i ,
-
where n (hij , </>ol is the projection operator which has value 1 if the induced
metric and matter fields on S have the given values and is zero otherwise.
One can cut the manifold M at the surface S to obtain a new manifold M
bounded by two copies S and S' of S. One can then define p(hij' c/> 0 ; h;j, cf>� ) to
be the path integral over all metrics and matter fields on M which· agree with
the given values hij• c/> 0 on S and h;j, cf>� on S. The quantity p can be regarded
as a density matrix describing the quantum state of the universe as seen from
a single spacelike surface for the following reasons :
(i) The diagonal elements of p, that is, when hij = h;j and cf> o = cf>� , give the
probability of finding a surface S with the metric hij and matter
fields cf> 0 .
(ii) If S divides M into two parts, the manifold M will consist of two
disconnected parts, M + and M The path integral for p will factorise :
_ .
p(hij • cf> o ; h;j, cf>�) = \!' + (hij • cf> o)\J' _ (h;j, cf>� ) ,
where the wave functions \!' + and \!' are given by the path integral
over all metrics and matter fields on M + and M respectively which
_
have the given values on S and S'. If the matter fields cf> are CP invariant,
_
\!' + = \!' and both are real (Hawking , 1985). \!' is known as 'The Wave
_
Function Of The Universe' . A density matrix which factorises can be

interpreted as corresponding to a pure quantum state.
(iii) If the surface S does not divide M into two parts, the manifold M will be
connected . In this case the path integral for p will not factorise into the
product of two wave functions. This means that p will correspond to the
density matrix of a mixed quantum state, rather than a pure state for
which the density matrix would factorjse (Page, 1986; Hawking, 1987).
)
One can think of the density matrices which do not factorise in the
following way : Imagine a set of surfaces I; which , together with S, divide the
spacetime manifold M into two parts. One can take the disjoint union of the
� and S as the surface which is used to define p (there is no reason why this
su rface has to be connected) . In this case the manifold M will be
disconnected and the path integral for p will factorise into the product of two
wave functions which will depend on the metrics and matter fields on two
sets of surfaces, S, I; and S' , r; . The quantity p will therefore be the density
matrix for a pure quantum state. However, an observer will be able to
measure the metric and matter fields only on one connected component of
the surface (say, S) and will not know anything about their values on the
other components, Ji, or even if any other components are required to divide
the spacetime manifold into two parts. The observer will therefore have to
sum over all possible metrics and matter fields on the surfaces 7;. This
summation or trace over the fields on the 1; will reduce p to a density matrix
corresponding to a mixed state in the fields on the remaining surfaces S and
638 S. W Hawking
S' . It is like when you have a system consisting of two parts A and B.
Suppose the system is in a pure quantum state but that you can observe only
part A. Then, as you have no knowledge about B, you have to sum over all
possibilities for B, with equal weight. This reduces the density matrix for the
system from a pure state to a mixed state.
The summation over all fields on the surfaces 1{ is equivalent to joining
the surfaces 1{ to r; and doing the path integral over all metrics and matter
fields on a manifold M whose only boundaries are the surfaces S and S' .
There is an overcounting because, as well as summing over all metrics and
matter fields, one is summing over all positions of the surfaces Ii in M .
However, the path integral over these extra degrees of freedom can be
factored out by introducing ghosts. The reduced path integral is then the
same as that for the density matrix p for a single pair of surfaces S and S' .
Thus one can see that the reason that the density matrix for S corresponds to
a mixed state is that one is observing the state of the universe on a single
spacelike surface and ignoring the possibility that spacetime may be not
simply connected and so require other surfaces Ii as well as S to divide M
into two parts.
14.4 The Wheeler-DeWitt equation

In a neighbourhood of the boundary surface S of the manifold M , one can
write the metric gµv in the (d - 1)+ 1 form :
ds 2 (N 2 + Ni NJ dt 2 + 2Ni dxi dt + hii dxi dxi
= ,
where S is the surface t = 0. The Euclidean action can then be written in the
Hamiltonian form :
l= - f
dt d' - ' x(n'ili;; + n0 <,6 - NH" - N ,H ' ) ,
i
where n i = - (h1 1 2/16n)(Kii _ hiiK) is the Euclidean momentum conjugate
to h ii ' Kii is the second fundamental form of S,
�
H o = 16 n G iiktn iinkt -
l n h i 1 2 d - 1 R + T oo
Hi = - 2nii; i + To i
G iik t = 1h - 1 1 2 (h ikhi, + huhjk - hiihk1 ) .
As was stated above, the components of gµ v that lie out of the su rface S can
be given any values by a diffeomorphism of M that leaves S fixed. This
means that the variational derivative of the path integral for p with respect
to N and Ni on S mu st be zero :
bp
bN .
I J
= - d [g µ v] d[ ¢ o]
bl
bN .
I
exp( - i) =
-
Hip =O
[ [
bp
bN J
= - d gµ vJ d </J oJ
bf
bN
-
exp( - I) = Hp = O,
-
where the operators fl and Hi are obtained from the corresponding classical
j
expressions by replacing the Euclidean momentum ni by - b/bhij and n</> by
- 6/6¢.
The first equation is called the momentum constraint. It is a first-order
equation for p on superspace, the space W of all metrics h ij and matter fields
¢ on a su rface S. It implies that p is the same for metrics and matter fields
which can be obtained from each other by coordinate transformations in S.
The second equation is called the Wheeler-DeWitt equation. It holds at
each point of superspace, except where h ii = h;i and ¢0 = ¢ � . When this is
true, the separation between S and S' in the metric gµ v on the manifold M
may be zero . In this case, it is no longer true that the variation of p with
respect to N is zero . There is an infinite dimensional delta function on the
right-hand side of the Wheeler-DeWitt equation . Thus, the Wheeler
DeWitt equation Is like the equation for the propagator, G(x, x') =
(¢(x)¢(x')) :
( - D + m 2 )G(x, x') = 6(x, x') .
As the point x tends towards x' , the propagator diverges like r 2 d , where r
-
is the distance between x and x'. Thus G(x , x') will be infinite. Similarly,
p(h ij, ¢0; hii' ¢0), the diagonal elements of the density matrix, will be infinite.
This infinity arises from Euclidean geometries of the form S x S1 , where the
S1 is of very short radius. However, we are interested really only in the
probabilities for Lorentzian geometries, because we live in a Lorentzian
universe, not a Euclidean one. One can recognise the part of the density
matrix p that corresponds to Lorentzian geometries by the fact tha� it will
oscillate rapidly as a function of the scale factor of the metrics hij and h ;i
(Hawking, 1984b). One therefore wants to subtract out the infinite,
Euclidean, component and leave a finite, Lorentzian, component. One way
of doing this is to consider only spacetime manifolds M which the surface S
divides into two parts. The density matrix from such geometries will be of
the factorised form :
p(h ii' ¢0 ; h ;i , ¢ � ) = 'P (h ii' </J o ) 'P (h ;i , ¢ � ),
where the wave function lJ' obeys the Wheeler-DeWitt equation with no
640 S. W. Hawking
delta function on the right-hand side. This part of the density matrix will
therefore remain finite when hu = h;i and ¢ 0 = ¢�. In a su persymmetric
theory, such as supergravity or superstrings, the infinity at the diagonal in
the density matrix would probably be cancelled by the fermions.
14.5 Minisuperspace
The Wheeler-DeWitt equation can be regarded as a second-order
differential equation for p or \JI on superspace , the infinite-dimensional
space of all metrics and matter fields on S. It is hard to solve such an
equation . Instead , progress has been made by using finite dimensional
approximations to superspace, called minisuperspaces, first introduced by
Misner ( 1970). In other words, one reduces the infinite number of degrees of
freedom of the gravitational and matter fields and of the gauge to a finite
number and solves the Wheeler-DeWitt equation on a finite-dimensional
space.
1 4 .5 . 1 de Sitter model
The simplest example is a homogeneous isotropic four-dimensional universe
with a cosmo�ogical constant and metric
ds 2 = o- 2 [N 2 dt 2 + a 2 dQ�] .
[ [
The action is
I
1 da 2
J J
1
I = -- dtNa - - + I - A.a 2
_
'
2 N 2 dt
where o- 2 = 1nm; , a is the radius of the 3-sphere space-like surfaces and
A. =-to- 2 A . One can choose N = a. The first two terms in the Euclidean action
are negative definite. This means that the path integral over a does not
converge. However, one can make the path integral converge by taking a to
be imaginary. This corresponds to integrating the conformal factor over a
contour parallel to the imaginary axis (Gibbons et al . , 1978).
With a imaginary, the action is the same as that of the anharmonic
oscillator. The density matrix p(a, a') is given by a path integral over all
values of a on a manifold M bounded by surfaces with radii a and a'. There
are two kinds of such manifold : ones that have two disconnected
components, which correspond to spacetimes that are divided in two by S,
and connected ones, which correspond to non-simply connected spacetimes
that S does not divide.
Consider fi rst the case in which S divides M in two. The density matrix
from these geometries that S divides into two is the product of wave
functions :
p (a, a') = \Jl(a)'I'(a'),
where the wave function 'I' is given by a path integral over compact 4-
geometries bounded by a 3-sphere of radius a or a' . One would expect this
path integral to be approximately A exp( B), where B is the action of a
-
solution of the classical Euclidean field equations with the given boundary
conditions and the prefactor A is given by a path integral over small
fluctuations about the solution of the classical field equations. The compact
homogeneous isotropic solution of the Euclidean field equations is a 4-
sphere of radius A. - 1 1 2 . A 3-sphere of radius a < A. - 1 1 2 can fit into such a 4-
sphere in two positions : it can bound more or less than half the 4-sphere.
The action B of both these solutions of the classical equations is negative,
with the action of more than half the 4-sphere being the more negative. One
might therefore expect that this solution would provide the dominant
contribution to the path integral . However, if one takes the scale factor a to
be imaginary, in order to make the path integral converge, and then
analytically continues back to real a, one finds that the dominant
contribution comes from the solution that corresponds to less than half the
4-sphere, rather than the other solution which corresponds to more than
half the 4-sphere, as one might have expected . This conclusion also follows
from an analysis of the path integral in the K repr�sentation. (Hartle and
Hawking , 1983).
In terms of the gauge choice N = a , used above, the path integral is over a
with a the given value at t = 0 and a = 0 at t = + oc . This path integral is the
same as that for the propagator for the an harmonic oscillator from ia at t = 0
to 0 at t oc . But this gives the ground state wave function. Thus
=
'I'(a) oc Re(A0(i a)),

where A0(x) is the ground state wave function of the anharmonic oscillator.
For small x, A0(x) behaves like exp( �x 2 ). Thus \Jl(a) behaves
exponentially like exp(�x 2 ). This agrees with the estimates from the action of
-
less than half the 4-sphere, as above. However, for a > A - 1 1 2 , there is no
Euclidean solution of the classical field equations for a compact
homogeneous isotropic 4-space bounded by a 3-sphere of radius a. Instead
there are complex metrics which are solutions of the field equations with the
required properties. Near the 3-sphere of radius a, one can take a section
through the complexified spacetime manifold on which the metric is real and
Lorentzian. This is reflected in the fact that A0(i a) will oscillate for a > A - 1 1 2 :
exponential wave functions correspond to Euclidean 4-geometries and
642 S. W. Haw king
oscillating wave functions correspond to Lorentzian 4-geometries

(Hawking , 1984b).
For large a, \J'(a) behaves like a - 1 cos(A. 1 '2 a 3 ). One can interpret this by
writing the wave function in the WKB form : C(exp(i S) + exp( - i S)), where
S is a rapidly varying phase factor and C is a slowly varying amplitude. The
wave function will satisfy the Wheeler-DeWitt equation to leading order if
the phase factor S obeys the classical Hamilton-Jacobi equation. Thus, an
oscillating wave function will correspond in the classical limit to an (n - 1)
dimensional family of solutions of the classical Lorentzian field equations,
where n is the dimension of the minisuperspace.
In the example above, n = 1 . The oscillating part of the wave function
corresponds to the classical de Sitter solution which collapses from infinite
radius to a minimum radius a = A. - 1 1 2 and then expands again exponentially
to infinite radius. The classical Lorentzian solution does not go below a
radius of A. - 1 12 , so one can interpret the exponentially damped wave
function below that radius as corresponding to a Euclidean geometry in the
classically forbidden region. Note that, for this explanation to make sense,
the wave function has to decrease with decreasing a, and not increase as
authors such . as Linde and Vilenkin have argued on the analogy of
tunnelling 'from nothing'. Anyway, if one believes that the quantum state of
the universe is determined by a path integral over compact geometries, one
has no freedom of choice of the solution of the Wheeler-DeWitt equation : it
has to be the one that increases exponentially with increasing a.
Another feature of the wave function that is worth remarking on is that it
is real. This means that , in the oscillating region, the WKB ansatz is
C(exp(i S) + exp( -i S)). One can regard the first term as representing an
expanding universe and the second a contracting universe. More generally ,
if the wave function represents some history of the universe, it also
represents the CP T image of that history (Hawking, 1985). This should be
contrasted with the approach of Vilenkin and others, who try to choose a
solution of the Wheeler-DeWitt equation which corresponds only to
expanding universes. The fallacy of this attempt is that the direction of the
time coordinate has no intrinsic meaning : it can be changed by a coordinate
transformation. The physically meaningful question is : how does the
entropy or degree of disorder behave during the histories of the universe that
are described by the wave function? The minisu perspace models considered
here are too simple to answer this but it will be discussed for models with the
full number of degrees of freedom in Section 14.7.
The contribution to the density matrix from geometries that S does not
divide into two parts is given by a path integral with a fixed at the given
values at t = 0 and t = ti for some Euclidean time interval ti . But this is equal
to the real part of the propagator K(i a, O; i a', ti) for the anharmonic
oscillator from i a at t = O to i a' at t = ti .
"
where A11(x) are the wave functions of the excited states of the anharmonic
oscillator and E11 are the energy levels. To obtain the density matrix one has
to integrate over all values of ti because the two surfaces can have any time
separation :
P (a ' a') = Re
f.00 K(i a , O , i a' , t i ) dti = Re "
· A 1 (i a)A 11(i a')
L... 1 ·
0 n
E
II
One can interpret this as saying that the universe is in the state specified by
i
the wave function Re(A11(i a)) with the relative probability (E") - . Note that
the universe need not be 'on shell' in the sense that the Wheeler-DeWitt
operator acting on A11 is not 0, but E11 • This term in the Wheeler-DeWitt
equation acts as if the universe contained a certain amount of negative
energy radiation . It will cause the classical solution corresponding to A " by
the WKB approximation to bounce at a larger radius than ;_ - i 1 2 . Thus, the
effect of the universe being in a mixed quantum state might be observable.
However, at large values of a, the effect of the negative energy radiation
would be very small and the universe would expand exponentially, like the
de Sitter solution.
14 .5 .2The massive scalar field model

The de Sitter model was interesting because it showed that the 'no
boundary' proposal for the quantum state of the universe leads to inflation if
there is some process which gives rise to an effective cosmological constant
in the early universe. However, the universe is not expanding exponentially
at the present time , so there has to be some way in which the cosmological
'constant' can reduce to zero at late times. One mechanism, and possibly the
only one, for generating such a decaying effective cosmological constant is a
scalar field with a potential which has a minimum at zero and which is
exponentially bounded . I shall consider the simplest example, a massive
scalar field .
The action of a homogeneous isotropic universe of radius a with a massive
644 S. W Hawkin g
[[ J [
scalar field ¢ that is constant on the surfaces of homogeneity is
_ 1
I [
I = -2 dtNa 2
N
1
JJ
d a 2 2 d¢ 2
dt
a
- dt J
+ 1 - a 2 m 2A'+', 2 .
Unfortunately, in this case, there does not seem to be any simple

prescription for making the Euclidean action positive definite. Taking a
imaginary leaves the kinetic term for ¢ negative, while taking ¢ imaginary
would cu re this problem but would make the mass term negative. One
could, however, make the action positive in this manner if the potential was
pure ¢4. On physical grounds, one would not expect that there would be a
qualitative difference between the behaviour of a universe in which the scalar
potential was ¢ 2 and one in which it was ¢4.
In the case that the surface S divides the spacetime into two parts, the
[
wave function will obey the Wheeler-DeWitt equation
1 1 a a 1 a1 ]
- - -aP- - - 2 - a 2 + a 4 m 2 ¢ 2 \f [a, ¢J = 0,
2 aP aa aa a 2 acp
--
where p reflects some of the uncertainty in the factor ordering of the

operators in the Wheeler-DeWitt equation. It is thought that the value of p
does not have much effect , so it is usual to take p = 1 , because this simplifies
the equation. One can introduce new coordinates :
x = a sinh ¢, y = a cosh ¢ .
I n these coordinates, the Wheeler-DeWitt equation becomes
[:;2 -::2 V} + '(x, y) = O,
where V = (y 2 - x 2 )[ - 1 + (y 2 - x 2)m 2 (arctanh x/y) 2] .

For small values of a , one can expect that \f is approximately A exp( - B) ,
where B is the action of a solution of the Euclidean field equations. If ¢ � 1
and a < 1/m¢, the value of ¢ will not vary much over the solution and the
m 2 ¢ 2 term in the action will act as an effective cosmological constant. One
would therefore expect B to be the action of the smaller part of a 4-sphere of
radius 1/m¢, bounded by a 3-sphere of radius a . From the de Sitter model ,
one would expect the wave function to oscillate for a > 1/m¢ and the phase
factor S to be �m ¢ a 3 , the analytic continuation of B. Such a wave function is
a solution to the Wheeler-DeWitt equation to leading order.
One can interpret the oscillating part of the wave function as
corresponding to a complex compact metric which is a solution of the field
equations and which is bounded by the surface S. In a neighbou rhood of S
one can take a section through the complexified spacetime manifold on
which the met ric is nearly real and Lorentzian. This solution will have a
minimum radius of order 1/rn¢ and will expand exponentially with ¢ slowly
decreasing . It will be a quantum realisation of the 'chaotic inflation' model
proposed by Linde ( 1983).
After an exponential expansion of the universe by a factor of order
exp {-!¢ 2 ), the scalar field will start to oscillate with frequency rn. The energy
momentum tensor of the scalar field will change from that of an effective
cosmological constant to that of pressure-free matter. The universe will
change from an exponential expansion to a matter-dominated one. In a
model with other matter fields, one would expect the energy in the massive
scalar field oscillations to be converted into zero rest mass particles. The
universe would then expand as a radiation-dominated model .
The universe would expand to a maximum radius and then recollapse.
One would expect that if such complex, almost Lorentzian , geometries
contributed to the wave function in their expanding phase, they would also
contribute in their contracting phase. However, although a few solutions
will bounce at small radius and expand again (Hawking , 1984b; Page,
1985a, b), most solutions will collapse to a singularity. They will give an
oscillating contribution to the wave function, even in the region a < 1/rn<P of
superspace where the dominant contribution is exponential . It will also
mean that the boundary condition for the Wheeler-DeWitt equation on the
light cone x = + y is not exactly '¥ = 1 , as was assumed in some earlier papers
(Hawking and Wu , 1985; Moss and Wright, 1983).
The density matrix from geometries that S does not divide into two parts
has not been calculated yet . By analogy with the de Sitter model , one might
expect that the part which corresponds to Lorentzian geometries would
behave like solutions with a massive scalar field and negative energy
radiation . One would not expect the negative energy to prevent collapse to a
singularity.
To summarise, in this model , the universe begins its expansion from a
non-singular state. It expands in an inflationary manner, goes over to a
matter or radiation-dominated expansion , reaches a maximum radius and
recollapses to a singularity. This will be discussed fu rther in Sections 14.7
and 14.8.
14.6 Beyond minisuperspace

The minisuperspace models were useful because they showed that the 'no
boundary' proposal for the quantum state of the universe can lead to a
universe like the one that we observe, at least in its large scale features .
646 S. W Hawking
However, ultimately one would like to know the density matrix or wave
function on the whole of superspace, not just a finite-dimensional subspace.
This is a bit of a tall order but one can use a 'midisuperspace' approximation
in which one takes the action to all orders in a finite number of degrees of
freedom and to second order in the remaining degrees of freedom.
A treatment of the massive scalar field model on these lines has been given
by Halliwell and Hawking ( 1985). The two degrees of freedom of the model
described above are treated exactly, and the rest as perturbations on the
background determined by the two-dimensional minisuperspace model . As
in the model above, the oscillating part of the background wave function
corresponds by the WKB approximation to a universe which starts at a
minimum radius, expands in an inflationary and then a matter-dominated
manner, reaches a maximum radius and recollapses to a singularity.
From the 'no boundary' condition the behaviour of the perturbations is
determined by a path integral of the perturbation modes over the compact
geometries represented by the background wave function. In the case of
Euclidean geometries that are part of a 4-sphere or of complex geometries
that are near such a Euclidean geometry, one can use an adiabatic
approximation. to show that the perturbation modes are in their ground
state, with the minimum excitation compatible ·with the uncertainty
principle. This means that the Lorentzian geometries that correspond to the
oscillating part of the wave function start off at the minimum radius with all
the perturbation modes in the ground state. As the universe inflates, the
adiabatic approximation remains good and the perturbation modes remain
in their ground states until their wavelength becomes longer than the
horizon size or, in other words, their frequency is red shifted to less than the
expansion time scale. After this, the wave functions of the perturbation
modes freeze and do not relax adiabatically to remain in the ground state as
the frequency of the modes changes.
The perturbation modes remain frozen until the wavelength of the modes
becomes less than the horizon size again during the matter- or radiation
dominated expansion. Because they have not been able to relax
adiabatically, they will then be in a highly excited state. After this, they will
evolve like classical perturbations of a Friedmann universe. They will have a
'scale free' spectrum, that is, their rms amplitude at the time the wavelength
equals the horizon size will be independent of the wavelength. The
amplitude will be roughly lO(m/mp), where m is the mass of the scalar field.
Thus they would have the right amplitude of about 10 - 4 to account for
galaxy formation if m is about 10 1 4 GeV.
In order to generate sufficient inflation, the initial value of the scalar field
¢ has to be greater than about 8 . However, with m = 10- smp , the energy
density of the scalar field will still be a lot less than the Planck density. Thus
it may be reasonable in quantum cosmology to ignore higher-order terms
and extra dimensions.
In the recollapse phase the perturbations will continue to grow classically.
They will not retu rn to their ground state when the universe becomes small
again, as I suggested (Hawking , 198 5). The reason is that when they start
expanding, the background compact geometry bounded by the surface S is
near to the Euclidean geometry of half a 4-sphere. On such a background the
adiabatic approximation will hold for the perturbation modes, so they will
be in their ground state. However, when the universe recollapses, the
background geometry will be near a Lorentzian solution which expands and
recontracts. The adiabatic approximation will not hold on such a
background . Thus the perturbation modes will not be in their ground state
when the universe recollapses, but will be highly excited .
14. 7 The direction of time

The quantum state defined by the 'no boundary' proposal is CPT invariant
(Hawking, 1985), though this is not true of other quantum states, such as
that proposed by Vilenkin ( 1 986) . Yet the observed universe shows a
pronounced asymmetry between the future and the past. We remember
events in the past but we have to predict events in the future. Imagine a tall
building which is destroyed by an explosion and collapses to a pile of rubble
and dust. If one took a film of this and ran it backwards, one would see the
rubble and dust gather themselves together and jump back into their places
in the building. One would easily recognise that the film was being shown
backwards because this kind of behaviour is never observed : we do not see
tower blocks jumping up . Yet it is not forbidden by the laws of physics.
These are CP T invariant . In fact, the laws that are important for the
structure of buildings are invariant under C and P separately. Thus, they
must be invariant under T alone. In other words, if a building can collapse, it
can also resurrect itself.
The explanation that is usually given as to why we do not see buildings
jumping up is that the second law of thermodynamics says that entropy or
disorder must always increase with time, and that an erect building is in a
much more ordered state than a pile of rubble and dust. However, this law
has a rather different status from other laws, such as Newton's law of
648 S. W Hawking
gravity. First, it is not an absolute law that is always obeyed : rather it is a

statistical law that says what will probably happen. Second, it �s not a local
law like other laws of physics : it is a statement about boundary conditions.
It says that if a system starts off in a state of high order, it is likely to be found
in a disordered state at a later time, simply because there are many more
disordered states than ordered ones.
The reason that entropy and disorder increase with time and buildings fall
down rather than jump up is that the universe seems to have started out in a
state of high order in the past. On the other hand, if, for some reason, the
universe obeyed the boundary condition that it was in a state of high order
at late times, then at earlier times it would be likely to be in a disordered state
and disorder would decrease with time. However, human beings are
governed by the second law and the boundary conditions, just like
everything else in the universe. Our subjective sense of the direction of time
is determined by the direction in which disorder increases because to record
information in our memories requires the expenditure of free energy and
increases the entropy and disorder of the universe. Thus, if disorder
decreased with time, our subjective sense of time would also be reversed and
we would still say that entropy and disorder increased with time. The second
law is almost a tautology : entropy and disorder increase with time because
we measure time in the direction in which disorder increases.
However, there remains the question of why should the universe have
been in a state of high order at one end of time? Why was it not in a state of
complete disorder or thermal equilibrium at all times? After all , that might
seem more probable as there are many more disorder states than order ones.
And why does the direction of time in which disorder increases coincide with
that in which the universe expands? Put it another way : why do we say that
the universe is expanding , and not contracting?
These questions can be answered only by some assumption on the
boundary conditions of the universe or, equivalently, on the class of
spacetime geometries in the path integral . As we have seen , the 'no
boundary' condition implies that the universe would have started off in a
smooth and ordered state with all the inhomogeneous perturbations in their
ground state of minimum excitation . As the universe expanded, the
perturbations would have grown and the universe would have become more
inhomogeneous and disordered. This would answer the questions above.
But what would happen if the universe, or some region of it , stopped
expanding and began to collapse? At first I thought (Hawking, 1985) that
Quantum cosmology 64S
entropy and disorder would have to decrease in the contracting phase so

that the universe would get back to a smooth state when it was small again .
This was because I thought that at small values of the radius a, the wave
function would be given just by a path integral over small Euclidean
geometries. This would imply \f = 1 on the light cone x = + y in the model
described above and that the adiabatic approximation would hold for the
perturbation modes , which would therefore be in their ground state.
However, Page (1985b) pointed out that there would also be a contribution to
the wave function from compact , complex , almost Lorentzian geometries
that represented universes that started at a minimum radius, expanded to a
maximum and recollapsed , as described above. This was supported by work
by Laflamme ( 1987), who investigated a minisuperspace model in which the
su rfaces 5 had topology 5 1 x 52. He also found almost Lorentzian solutions
which started in a non-singular manner but recollapsed to a singularity. The
adiabatic approximation for the perturbation modes would not hold in the
recollapse. Thus they would not return to their ground states , but would get
even more excited as the collapse continued. The universe would get more
and more inhomogeneous and disorder would continue to increase with
time.
There remains the question of why we observe that the direction of time in
which disorder increases is also the direction in which the universe is
expanding . Because the 'no boundary' quantum state is CPT invariant,
there will also be histories of the universe that are the CPT reverses of that
described above. However, intelligent beings in these histories would have
the opposite subjective sense of time. They would therefore describe the
universe in the same way as above : it would start in a smooth state, expand
and collapse to a very inhomogeneous state . The question therefore
becomes : why do we live in the expanding phase? If we lived in the
contracting phase, we would observe entropy to increase in the opposite
direction of time to that in which the universe was expanding. To answer
this, I think one has to appeal to the weak anthropic principle. The
probability is that the universe will not recollapse for a very long time
(Hawking and Page, 1986). By that time, the sta rs would all have burnt out
and the baryons would have decayed . The conditions would therefore not be
su itable for the existence of beings like us. It is only in the expanding phase
that intelligent beings can exist to ask the question : why is entropy
increasing in the same direction of time as that in which the universe is
expanding?
650 S. W. Hawking
14.8 The origin and fate of the universe

Does the universe have a beginning and/or end?
If the 'no boundary' proposal for the quantum state is correct, spacetime is
compact. On a compact space, any time coordinate will have a minimum
and a maximum. Thus, in this sense, the universe will have a beginning and
an end.
Will the beginning and end be singularities?
Here one must distinguish between two different questions : whether there
are singularities in the geometries over which the path integral is taken, and
whether there are singularities in the Lorentzian geometries that correspond
to the density matrix by the WKB approximation . A singularity cannot
really be regarded as belonging to spacetime because the laws of physics
would not hold there. Thus, the requirement of the 'no boundary' proposal
that the path integral is over compact geometries only rules out the existence
of any singularities in this sense. Of course, one will have to allow compact
metrics that are not smooth in the path integral , just as in the integral over
particle histories one has to allow particle paths that are not smooth but
satisfy a Hold�r continuity condition . However, one can approximate such
paths by smooth paths. Similarly, in the path integral for the universe, it
must be possible to approximate the non-smooth metrics in a suitable
topology by sequences of smooth metrics because otherwise one could not
define the action of such metrics . Thus, in this sense, the geometries in the
path integral are non-singular.
On the other hand, the Lorentzian geometries that correspond to the
density matrix by the WKB approximation can and do have singularities . In
the minisuperspace model described above, the Lorentzian geometries
began at a non-singular minimum radius or 'bounce' and evolve to a
singularity in general , in the direction of time defined by entropy increase. I
would conjecture that this is a general feature : oscillating wave functions
and Lorentzian geometries arise only when one has a massive scalar field
which gives rise to an effective cosmological constant and Euclidean
solutions which are like the 4-sphere. The Lorentzian solutions will be the
analytic continuation of the Euclidean solutions. They will start in a smooth
non-singular state at a minimum radius equal to the radius of the 4-sphere
and will expand and become more irregular. When and if they collapse, it
will be to a singularity.
One could say that the universe was 'created from nothing' at the
minimum radius (Vilenkin , 1982) . However, the use of the word 'create'
Quantum cosmology 65 1
would seem to imply that there was some concept of time in which the
universe did not exist before a certain instant and then came into being. But
time is defined only within the universe, and does not exist outside it, as was
pointed out by Saint Augustine (400) : 'What did God do before He made
Heaven and Earth? I do not answer as one did merrily : He was preparing
Hell for those that ask such questions. For at no time had God not made
anything because time itself was made by God .'
The modern view is very similar. In general relativity, time is just a
coordinate that labels events in the universe. It does not have any meaning
outside the spacetime manifold . To ask what happened before the universe
began is like asking for a point on the Earth at 9 1 ° north latitude; it just is
not defined . Instead of talking about the universe being created, and maybe
coming to an end , one should just say : The universe is.
References
Gibbons, G . W., Hawking, S. W. and Perry, M . J. ( 1978). Nucl. Phys., Bl38, 14 1 .
Goroff, M . H . and Sagnotti, A . ( 1985) . Phys. Lett., 160B, 8 1 .
Halliwell, J . J . and Hawking, S . W. ( 1985). Phys. Rev., 031, 1 777.
Hartle, J . B . and Hawking, S . W . ( 1983). Phys. Rev., 028, 2960.
Hawking , S. W. ( 1982) . In Astrophysical Cosmology. Proceedings of the Study Week on
Cosmology and Fundamental Physics, ed . H . A. Bruck, G. V. Coyne and M. S.
Longair. Pontificia Academiae Scientarium : Vatican City.
Hawking, S. W. and Penrose, R. ( 1970). Proc. Roy. Soc. Lon., A314, 529.
Hawking, S. W. and Ellis, G. F. R. ( 1973) . The Large Scale Structure of Space- Time.
Cambridge University Press: Cambridge.
Hawking, S. W. ( 1984a) . Phys. Lett., 150B , 339.
Hawking , S. W. ( 1984b). Nucl. Phys. , B239, 257.
Hawking , S . W. and Wu , Z. C. ( 1985) . Phys. Lett. , 151B, 15.
Hawking, S . W. ( 1985). Phys. Rev., 032 , 2489.
Hawking, S. W. and Page, D. N . ( 1986). Nucl. Phys., B264 , 1 8 5 .
Hawking, S. W. ( 1987). Physica Scripta (in press).
Laflamme, R. ( 1987) . The wave function of a S1 x S2 universe. Preprint , to be published.
Lifshitz, E. M . and Khalatnikov, I . M . ( 1963). Adv. Phys., 12, 185.
Linde, A. D. ( 1 98 3) . Phys. Lett., 1 29B, 177.
Linde, A . D. ( 1985). Phys. Lett., 162B, 2 8 1 .
Misner, C . W. ( 1 970) . I n Magic without Magic , ed . J. R . Klauder. Freeman : San
Francisco.
Moss, I . and Wright, W. ( 1983). Phys. Rev., 029, 1067 .
Page, D. N. ( 1985a). Class. & Q .G., 1 , 4 1 7.
'
Page, D . N . ( 1985b). Phys. Rev., 032, 2496 .

Page, D. N . ( 1986) . Phys. Rev . , 034, 2267 .
Rubakov, V. A . , Sazhin, M . V . and Veryaskin, A . V . ( 1982). Phys. Lett., 1 15B, 189.
Saint Augustine (400). Confessions. Re-edited in Encyclopedia Britannica ( 1952).
Vilenkin , A. ( 1982) . Phys. Lett . , 1 17B , 25.
Vilenkin, A. ( 1 986). Phys. Rev., 033, 3560.
15
Superstring unification t
J O H N H . S C H W A RZ !
15. 1 Introduction
Superstring theory is an approach to the unification of fundamental
particles and forces that has attracted a great deal of attention in the last two
years [ 1] . It offers the possibility of overcoming many of the shortcomings of
the standard model and providing a unified theory free from much of the
arbitrariness that. is inherent in conventional point-particle theories.
The standard model of electroweak and strong forces has enjoyed an
enormous amount of success. Indeed, it appears to be consistent with all
established particle physics experiments. This being the case, the first
question one should ask is why one should even be looking for something
better. Most criticisms of the standard model are based on the fact that it
requires a number of arbitrary choices and fine-tuning adjustments of
parameters. These features do not prove that it is wrong or even incomplete.
However, given the history of successes in elementary particle physics, it is
natural to seek a deeper underlying theory that can account for many of the
arbitrary choices and parameters. These include the choice of gauge groups
and representations, the number of families of quarks and leptons, the
origins of the Higgs symmetry-breaking mechanism, and the specific values
of various parameters.
The aesthetic shortcomings of the standard model or a grand unified
model do not prove that it is wrong or incomplete. The thing that does this
most convincingly is the absence of gravity. There is a straightforward way
to couple Einsteinian gravity to any relativistic field theory by following the
t Work supported in part by the US Department of Energy under contract DEAC 03-
8 1- ER40050.
:j: Based on a lecture presented at the Twenty-third International Conference on High
Energy Physics in Berkeley, California, July 1986.
Superstring unification 653
dictates of general coordinate invariance, and at the classical level it poses

no difficulties. However, the quantum theory, which is renorrnalizable
before gravity is appended, inevitably acquires non-renormalizable
ultra violet divergences. Many years of study strongly suggest that this is a
generic feature of all quantum field theories based on point particles . Even
Einsteinian gravity by itself has been shown to be singular at two loops [2] .
Pu re supergravity theories in four dimensions are finite at two loops, but are
very likely to be singular at three loops. The addition of extra spatial
dimensions (such as in eleven-dimensional supergravity) makes the
divergences even more severe, and therefore represents a step in the wrong
direction. This is where superstring theory comes to the rescue.
String theory was developed in the late 1960s and early 1970s in an
attempt to describe the strong nuclear force [3] . The first string theory was
plagued by a number of unrealistic features : the absence of fermions, the
presence of a tachyon and massless modes, and the necessity of 26-
dimensional spacetime. In an attempt to do better a second string theory
was introduced in 197 1 [4] . It incorporates a spectrum of fermions that was
proposed by Pierre Ramond [5] . In the form first discussed this theory also
contained a tachyon . It possesses two-dimensional superconformal
symmetry on the string world sheet but not spacetime supersymmetry. Five
years later, Gliozzi, Scherk , and Olive proposed a truncation of the
spectrum that eliminated the tachyon [6] . They noted that then there are an
equal number of bosons and fermions at each mass level and conjectured
that this version of the theory possesses spacetime supersymmetry. This
result was proved in 1980 by Michael Green and myself [7] . This theory is
called type I superstring theory in the modern classification (see Section
1 5.2). It was discovered in 1972 that superstring theories required ten
dimensional spacetime for their consistency [8] .
Neither the bosonic string theory nor superstring theory succeeded in
giving a realistic description of hadron physics. The extra dimensions were
an embarrassment as were the massless modes, which had no counterparts
in the hadronic spectrum. In 1974 Joel Scherk and I demonstrated that the
massless spin-2 mode that is present in both theories interacts in the
standard Einsteinian way at low energies [9] . (This was also shown by
Yoneya [ 10] .) We therefore suggested that it be interpreted as a graviton
rather than a hadron and that string theory be used as a basis for
constructing a unified description of gravity and all the other forces rather
than as a theory of hadrons. In other words, it was proposed that elementary
654 J. H. Schwarz
particles are strings rather than points as in conventional quantum field

theory.
The idea of using strings for unification offered a number of appealing
features. First of all, the existence of gravity would be understood as
necessary since it is a generic feature of all consistent string theories. Stri ng
theories do not possess ultraviolet divergences of the conventional type, and
therefore it appeared plausible that this type of theory could overcome the
divergence problem discussed above . To state the case as forcefully as
possible, string theory requires the existence of gravity, whereas the point
particle theories require that it does not exist. This conviction sustained my
enthusiasm for the subject throughout the ten-year peri od before its
widespread acceptance.
The use of string theory for unification rather than hadronic physics has a
number of other advantages. The extra dimensions can acquire a sensible
interpretation, because spacetime geometry is dynamical in a gravity theory
and there is a possibility that the unobserved dimensions could be
spontaneously compactified . Also , string theories do not possess adjustable
dimensionless parameters, nor much freedom to choose groups and
representations. So, if a scheme of this type is successful, it should be much
more predictive than the standard model . To put it diff�rently, the consistent
introduction of gravity in a unified theory is such a severe constraint that it
could have dramatic predictive consequences for low-energy physics.
String theory contains a fundamental length parameter, which is the
characteristic size of a string. In order for the strength of gravity to emerge
with the usual Newtonian value , it is necessary to associate this length with
the Planck length : a few times 10 - 33 cm. Thus in the unification context the
strings are taken to be some twenty orders of magnitude smaller than they
were in the hadronic context.
15.2 Classification of string theories

Strings have two possible topologies : open (with free ends) or closed (a loop
without ends). The most interesting theories consist of oriented closed
strings only. Type I superstring theory, which was the first one to be
understood , consists of unoriented open and closed strings.
How many string theories are there? The answer to this question is not yet
known. Ideally it will turn out that there is just one. This could happen if the
ones that are presently known are either inconsistent or equivalent to one
another. In order for a theory to be consistent it must be free from tachyons
(states of negative mass squared) and ghosts (states of negative norm). It is
also necessary that loop amplitudes be free from anomalies and possess
modular invariance : properties to which we shall return . Additional criteria,
perhaps of non-perturbative origin , may eventually be found to be necessary
also.
Six string theories are known that satisfy the consistency properties
mentioned above. Each of them requires that spacetime has ten dimensions
(nine space and one time) and that the two-dimensional world-sheet action
has superconformal symmetry. These theories are listed in Table 1 5 . 1 . There
is some evidence that the three heterotic theories are actually different
phases of the same theory, which would reduce the number of different
theories to be considered .
Superstring theories possess not only two-dimensional superconformal
symmetry but local ten-dimensional super-Poincare symmetry, a possibility
first pointed out by Gliozzi , Scherk and Olive [6] . There are three
possibilities for D = 10 supersymmetry, each of which can be realized in a
superstring theory. The minimal irreducible spin or in D = 10 satisfies
simultaneous M ajorana and Weyl properties and has 16 independent real
components. A theory with a single conserved Maj orana-Weyl supercharge
has N = 1 supersymmetry, a possibility realized by three of the six theories
listed in the table. There are two distinct possibilities for theories with two
conserved Majorana-Weyl supercharges. Either they have opposite
chirality (type IIA) or they have the same chirality (type JIB). This is the
maximum amount of supersymmetry that is possible in an interacting
theory (corresponding to N = 8 supersymmetry in four dimensions). The
type II theories are not promising for phenomenology because they do not
accommodate elementary Yang-Mills fields. However, the type JIB theory
is in some respects the most beautiful of all the theories, and it is not totally
out of the question that it could form the basis of a realistic phenomenology.
Table 1 5 . 1 . Six string theories

D = lO Yang-Mills
Name su persymmetry symmetry
Type I [1 1] N= l S0(32)
Type IIA [12] N=2
Type IIB [12] N=2
Heterotic [ 13] N= l S0(32)
Heterotic [ 13] N= l Ea x Ea
Heterotic [14] N=O S0(16) x S0( 16)
656 J. H. Schwarz
An important fact about the standard model is the left-right asymmetry in

the classification of quark and lepton multiplets (chirality) . The most
natural way to achieve this chiral asymmetry, starting from a theory in more
than four dimensions, is for the higher-dimensional theory to be left-right
asymmetric itself. This property is shared by all the theories in the table
except for the type IIA.
Another important consideration in choosing among the theories is a
desire to understand the origin of the hierarchy of mass scales. In particular,
one wants to understand why radiative corrections to the mass of Higgs
scalars do not destroy the enormous ratio between the electroweak scale and
the unification scale. The most promising possibility seems to be 'low
energy' supersymmetry spontaneously broken around the electroweak
scale. This can be realized in superstring theories with spacetime
supersymmetry.
In addition to the string theories listed in the table, there are some others
that are consistent in all respects except for the presence of a tachyon in the
spectrum. The original D 26 bosonic string is an example of such a theory,
=
and a number of others have been discovered recently [ 1 5] . It is an open

question whether any of these theories can be given a consistent
interpretation by identifying a stable ground state that is free from tachyonic
modes. Even if this is possible, one might have to pay an unacceptable price
in terms of a large cosmological constant. But that is a threat to all theories
once supersymmetry is broken.
15.3 Feynman diagrams
In conventional field theory, Feynman diagrams correspond to the distinct
topological possibilities for connecting the world lines of elementary
particles. The situation is analogous in string theory. Since strings are one
dimensional they sweep out a two-dimensional surface in spacetime. Thus
Feynman diagrams correspond to the topologically distinct world sheets
representing the possible spacetime histories of interacting strings. In
thinking about these surfaces it is extremely convenient to imagine carrying
out a Wick rotation and regarding the surfaces as being embedded in a space
of Euclidean metric. This makes the corresponding path integrals well
defined. This rotation can probably be rigorously justified by arguing that it
gives a unique unitary S matrix.
The classification of string Feynman diagrams is especially simple for the
theories that consist of oriented closed strings only. In these theories there is
a single fundamental string interaction, depicted in Fig. 1 5 . 1 , that describes
one string breaking into two , or two strings joining to give a single one. The
surface shown is topologically like a pair of pants. Since the surface is
smooth the particular spacetime point at which the interaction takes place is
a frame-dependent question . If the time slices are taken using planes that are
tilted relative to the ones shown , a different interaction point would be
identified . Thus there is no preferred point on the surface and the existence of
interaction is an inevitable feature of geometric origin, not something that
needs to be appended to the theory in an ad hoc way.
In calculating S-matrix elements one should consider amplitudes for
scattering closed strings that propagate to time + oo , i .e., with tubes
emerging from the surface that extend to infinity. However, the conformal
symmetry of the underlying two-dimensional field theory [ 16, 17] , implies
that world sheets that differ by a conformal transformation are equivalent .
In particular, one may map the asymptotic strings to finite coordinates.
When this is done, they are represented by points on the world sheet .
Altogether, the amplitude is then represented by a closed oriented surface
with 'punctures' representing the initial and final string states.
The topological classification of closed oriented two-dimensional surfaces
is characterized by a single integer, the genus g, which can be thought of as
the number of handles that is attached to a sphere. Thus, as shown in Fig .
1 5.2, the sphere has g = 0, the torus has g 1 , and so forth.
=
In the type I superstring theory the topological classification of diagrams

is a good deal more complicated. Since the strings themselves can be open or
closed and have no intrinsic orientation , the Feynman diagrams can have
boundaries and need not be orientable. This makes the analysis of this
theory more difficult. An important advantage of orientability is that an
Fig. 1 5 . 1 . A piece of world sheet in the shape of pants. The slice at time t 1
shows two closed strings while the one at time t 2 shows one closed string.
658 J. H. Schwarz
orientable surface can be regarded as a Riemann surface and powerful

techniques of complex analysis can be utilized.
In the type II and heterotic string theories, the topology of a diagram is
characterized by the genus g, which can be thought of as the number ofloops
in the diagram. Thus there is just a single diagram at each order of the
perturbation expansion. This is a remarkable simplification compared with
conventional field theories in which the number of diagrams at n loops is
roughly of order n ! . It has been shown by explicit computation that the one
loop amplitudes are finite [ 12 , 13] . Arguments have been made to suggest
that the loop amplitudes should be finite at every order, but this still requires
a careful analysis to be definitively established . This is likely to be settled
within a year or so . Assuming that the expected result emerges, we shall then
be in the happy position of having the first examples of perturbatively finite
quantum theories of gravity.
The analysis of multiloop (genus g > 1) string amplitudes i s a very rapidly
developing subject in which much beautiful work has been done in the last
year [ 18] . The details require state-of-the-art methods in the theory of
Riemann surfaces and algebraic geometry . All I can realistically hope to do
here is to sketch some of the basic ideas.
The loop amplitudes are given by integrals that represent a sum over all
conformally inequivalent geometries of the given topologies. Fortunately ,
these are finite-dimensional integrals. 3g - 3 + n complex parameters
describe the integration space M9 for a genus g amplitude with n external
particles. (This result is due to Riemann.) The structure of an amplitude is
Fig. 1 5.2. The topology o f closed orientable two-dimensional surfaces is

given by the genus g . The cases g = 0, 1, 2 are illustrated .
g=O 0 Sphere
g= l G Torus
g=2 � etc.
then given by an expression of the form
f,M.
dµ n (F ii)k , · k1 .
i <j
(1)
In this expression the indices i and j refer to the external particles and the k;
are the momenta (assumed to be on-shell). The function log Fij is
proportional to the Green's function on the world sheet between the
coordinates of particles i and j. Also, dµ represents a suitable integration
measure on the 'moduli space' M9 •
Not only the two-dimensional world sheet, but also the 2(3g - 3 + n)
dimensional moduli space has a complex structure. As a result it is possible
to express the measure dµ and functions F ii i n terms of holomorphic
functions on M9 • The space M9 is naturally expressed as a quotient of a space
r;,, known as Teichmiiller space, divided by an infinite discrete group of
transformations. The expressions dµ and Fii can be expressed as products of
various terms that are well defined on I;, but are multivalued ('line bundles')
on M9 • In order for the loop amplitude to be well defined it is necessary that
when the relevant products are formed, expressions that are single-valued
on M9 ('modular invariant') result. The fact that this works for each of the
various string theories is highly non-trivial . In fact, it almost completely
determines the various functions!
Moduli space M 9 has a very rich and subtle topology. In fact, it even has
boundary components that correspond to singular limits of the geometry in
which the world sheet 'degenerates'. The relevant issue for finiteness is
whether or not the measure dµ diverges on any of these boundary
components so fast as to give a singular integral . There are two distinct ways
in which the su rface can degenerate. One, depicted in Fig . 1 5.3(a), involves
the formation of a long thin tube that separates the diagram into two pieces.
The second, shown in Fig. 15. 3(b), involves the formation of a long thin tube
that does not separate the surface.
15.4 String field theory

The history of string theory has many strange aspects. One of them is that
the Feynman diagrams that I have been discussing were formulated before a
field theory that gives rise to them. The development of string field theory
was carried out in the light-cone gauge for bosonic strings in the 1970s [ 1 9]
and for superstrings in the early 1980s [20] . However, what one really wants
is a covariant gauge-invariant action that makes manifest all of the beautiful
underlying symmetries of the theory in a way that is not tied to a particular
660 J. H. Schwarz
choice of background fields. In short, one would like to know the string
analog of the Hilbert-Einstein action of general relativity. This has not yet
been achieved, but an enormous effort by many workers over the past year
has brought us much closer to this goal . In particular, a beautiful field theory
for open strings, correct at least at tree level , has been achieved . It is almost
impossible to give adequate credit to all the workers who have made
important contributions. Some of them are Siegel and Zwiebach [2 1] ,
Banks and Peskin [22] , Neveu and West [23] , Hata, ltoh, Kugo,
Kunitomo and Ogawa [24] . I will give a brief description of the results in
the form developed by Witten [25] .
One important remark is that in field theory an individual Feynman
diagram is given by an integral whose integration region is topologically
trivial : some sort of hypercube. Thus, since moduli space Mg has a
complicated topology, it cannot arise from a single diagram . What happens
is that many different field theory diagrams give different topologically
trivial pieces of the integration region Mg, all with the same integrand, and
the sum gives the complete genus g amplitude. Thus the individual diagrams
provide a 'triangulation' of moduli space [26] .
Witten's description of the field theory of open bosonic strings has many
analogies with Yang-Mills theory. This is not really surprising in as much as
open strings are an infinite-component generalization of Yang-Mills fields.
Fig. 15.3. The boundary of moduli space corresponds to singular limits in

which the surface 'degenerates'. This can occur by the formation of a long
thin tube that splits the diagram as in (a) or that does not split it as in (b).
(a)
(b)
It is pedagogically useful to emphasize these analogies in describing the

theory. The basic object in Yang-Mills theory is the vector potential A:(xP),
where µ is a Lorentz index and a runs over the generators of the symmetry
algebra. By contracting with matrices (A.0)ap that represent the algebra and
differentials dx µ we can define
(2a)
a, µ
a matrix of one form. This is a natural quantity from a geometric point of
view. The analogous object in open-string field theory is the string field
A [xP (a) , c(a)] . (2b)
This is a functional field that creates or destroys an entire string with
coordinates xP(a), c(a), where the parameter a is taken to have the range 0 :::;;
a :::;; n. The coordinates c(a) are anticommuting ghost degrees of freedom
that arise in the first quantization of the action. They are essential so that,
when A is expanded in an infinite sequence of point fields, there is an
appropriate set of auxiliary Stuckel berg fields at each mass level. The details
of such an expansion are quite complicated, but the mathematics of the
complete string field is not so bad. The Yang-Mills field is one of the infinity
of terms in such an expansion .
The string field A can be regarded as a matrix (in analogy to Aap) by
regarding the coordinates with 0 :::;; a :::;; n/2 as providing the left matrix index
and those with n/2 :::;; a :::;; n as providing the right matrix index as shown in
Fig. 15.4(a). One could also associate quark-like charges with the ends of the
Fig. 15.4. An open string has a left-hand (u < n/2) and right-hand (u > n/2)
segment, depicted in (a), which can be treated as matrix indices. The
multiplication A*B = C is depicted in (b).
L R
o=O 0 = 7r/2 0 = 7r
(a)
_JL
(b)
662 J. H. Schwarz
strings which would then be included in the matrix labels as well. This is a
minor and inessential complication, which we shall suppress. By not
including such charges we are constructing the string generalization of U( l)
gauge theory. U(l) gauge theory (without matter fields) is a free theory, but
the string extension has non-trivial interactions, as we shall see.
In the case of Yang-Mills theory, we can multiply two fields by the rule
(3a)
)'
This is a combination of matrix multiplication and antisymmetrization of

the tensor indices (the wedge product of differential geometry). This
multiplication is associative but non-commutative. A corresponding rule for
string fields is given by a * product,
A * B = C. (3b)
This infinite-dimensional matrix multiplication is depicted in Fig. 15.4(b).
One identifies the coordinates of the right half of string A with those of the
left half of string B and functionally integrates over them. This leaves string
C consisting of the left half of string A and the right half of string B. It is also
necessary to i_nclude a suitable factor involving the ghost coordinates at the
midpoint a = n/2.
A fundamental operation in gauge theory is exterior differentiation
A d A . In terms of components
--+
d A =!(oµ Av - ovAµ) dxµ j\ dxv ,

which contains the abelian field strengths as coefficients. Exterior
differentiation is a nilpotent operation , d2 = 0, since partial derivatives
commute and vanish under antisymmetrization. The non-abelian field
strength is given by the matrix-valued two-form
F = d A + A j\A, (4a)
or in terms of tensor indices,
Fµ v = oµ A v - ovAµ + [Aµ , A v] .
Let us now construct analogs of d and F for the string field. The operator
that plays the roles of d is the BRST operator Q. Q is a conserved fermionic
charge that arises as a consequence of a global fermionic symmetry of the
gauge-fixed quantum action with the ghost fields. The occurrence of this
BRST symmetry is a fundamental feature of gauge theories. Classically Q is
nilpotent (vanishing Poisson bracket with itself) for any spacetime
dimension. Quantum mechanically, Q 2 = 0 is satisfied only for D = 26 in the
case of bosonic strings [27] . Since this is an essential requirement, we must

make this choice. Q can be written explicitly as a differential operator
involving the coordinates X (a) , c(a) , but I shall not write out the formula
here. Given the operator Q , there is an obvious formula for the string theory
field strength , analogous to the Yang-Mills formula, namely
F = QA + A * A. (4b)
An essential feature of Yang-Mills theory is gauge invariance.
Infinitesimal gauge transformation can be described by a matrix of
infinitesimal parameters A(xP), which are functions of the spacetime
coordinates xP. The transformation rules for the potential and the field
strength are then
c5A = dA + [A , A] (5 a )
and
c5F = [F, AJ . (6a)
We can write down completely analogous formulas for the string theory,
namely
c5A = QA + [A, A] (5b)
and
c5F = [F, A] . (6 b)
In this case [A , AJ means A * A - A * A , of course. Since the infinitesimal
parameter A [xP(a), c(a)] is a functional , it can be expanded in terms of an
infinite number of ordinary functions. Thus the gauge symmetry of string
theory is infinitely richer than that of Yang-Mills theory, as required for the
consistency of the infinite spectrum of high spin fields contained in the
theory.
The next step is to formulate a gauge-invariant action. The key ingredient
in doing this is to have a suitably defined integral . In the case of Yang-Mills
theory we must integrate over spacetime and take a trace over the matrix
indices. Thus it is convenient to define J X as J d 4x Tr(X). In this notation
the usual Yang-Mills action is
f
I � gµ'g " F ,F,,,
µ (7)
which is easily seen to be gauge invariant. Requiring I to be stationary gives

the classical equations of motion Dµ F v = 0, where Dµ is a covariant
µ
derivative. Since F is quadratic in A, I describes cubic and quartic
interactions.
The definition of integration appropriate to string theory is a 'trace' that
664 J. H. Schwarz
identifies the left and right segments of the string field Lagrangian. Thus in
the case of string theory we define
Jx = J L u< 1t/2
3 2
b(x (u) - x(n - u)) · . · e - < i/ 2 )1/>(1tf > x . (8)
As indicated in Fig. 15.5(a), this identifies the left and right segments of X . A
ghost factor has been inserted at the midpoint. (c/> is a bosonized form of the
ghost coordinates.) The · · · signifies that analogous integrations should also
be performed for the ghost coordinates. We now have the necessary
ingredients to write a string action . If we try to emulate the Yang-Mills
formula we run into a problem, namely no analog of the metric gµp has been
defined. Rather than trying to find one, it proves more fruitful to look for a
gauge-invariant action that does not require one. The simplest possibility is
given by the Chem-Simons form
I
I - A • QA + iA . A . A . (9)
In the context of ordinary Yang-Mills theory the integrand is a three-form

and therefore such a term can only be introduced in three dimensions, where
it is interpret�d as giving mass to the gauge field. In string theory the
interpretation is different and the formula makes perfectly good sense. In
fact , it gives rise to the deceptively simple field equation F 0.
The fact that the string equation of motion is F 0 does not mean the = =
theory is trivial . If we drop the interaction term the equation of motion for
Fig. 1 5 . 5 . Integration ofa string functional requires identifying the left-hand

and right-hand halves as depicted in (a) . The three-string vertex , shown in (b) ,
is based on two multiplications and one integration and treats the three
strings symmetrically.
a=O
-------:
a = 7r/2 c
______,. a = 7r
(a)
R L
L�R
�R L"'\
(b )
the free theory is QA = 0, which is invariant under the abelian gauge

transformation bA = QA since Q 2 = 0. Once one imposes suitable
restrictions on ghost number, one can show this precisely reproduces the
known spectrum of the bosonic string.
The cubic string interaction is depicted in Fig. 15.5(b). Two of the segment
identifications are consequences of the * products in A * A * A and the third
is a consequence of the integration resulting in a symmetric expression. As
an alternative way of thinking about the interaction we can attach some flat
world sheet corresponding to free propagation of each of the three strings in
the interaction. This is depicted in Fig. 15.6. The folds in the drawing do not
imply intrinsic curvature of the surface and therefore have no physical
significance. (They are introduced for ease of depiction only.) There is
cu rvature at the interaction point, however. To see this, imagine drawing a
small circle of radius r around it. It has circumference 3nr (a contribution of
nr coming from each side of the surface). Thus the surface is flat everywhere
except at this one point where the curvature has a b function singularity.
A standard theorem implies that a closed orientable two-dimensional
surface of genus g has an integrated curvature proportional to x = 2(1 - g).
Thus, except for genus 1 , the surface does not admit a flat metric. In string
theory one is only interested in equivalence classes of metrics that are related
by conformal mappings. It is always possible to find representatives of each
equivalence class in which the metric is flat everywhere except at isolated
points where the curvature is infinite. Such a metric describes a surface with
conical singularities, which is not a manifold in the usual sense. In fact, it is
an example of a class of surfaces called orbifolds. The string field theory
construction of the amplitude automatically chooses a particular metric
which , as we have indicated, is of this type. It is also possible to choose
constant curvature metrics, but they do not arise from the string field theory.
Fig. 15.6. The three-string vertex with external string propagators attached
is given by a world sheet that is flat everywhere except at the interaction point
where a small circle of radius r has circumference 3nr.
666 J. H. Schwarz
The formation of string field theory described above is certainly very

beautiful. It is also deceptively simple. To really understand it in detail one
must define very carefully the various functional integrations that appear.
One way of doing this is to expand the fields in a Fock-space basis using an
infinite number of harmonic oscillators corresponding to the normal modes
of xP (a ) and c (a ). The functional integrations can then be carried out
explicitly leaving a Fock-space expression for the interaction vertex that can
be evaluated for any three states of the string spectrum . This calculation has
been carried out [28] , demonstrating that the formula is in fact well defined
and possesses the properties that it should. Evidence has also been found
that it reproduces the standard scattering amplitudes [29] .
15.5 Anomalies
The gauge invariance of classical gauge theories implies the existence of
conserved gauge currents, whose form can be deduced by the standard
Noether procedure. In Yang-Mills theories the associated conserved
charges are the symmetry generators. In general relativity the conserved
current is the · energy-momentum tensor and the associated charges are the
energy and momentu m . The classical conservation of these currents can ,
under certain circumstances, be destroyed by quantum effects called
anomalies. When this happens unphysical modes of the gauge fields become
coupled in the S matrix leading to a breakdown of unitarity and causality.
Thus, in general, all anomalies in gauge currents must cancel or else the
theory must be rejected as inconsistent .
The Feynman diagrams that give rise to anomalies typically arise at one
loop order and involve a loop of chiral fermions. In the well-known case of a
gauge theory in four dimensions the simplest diagram that can give rise to
anomalies is a triangle diagram, as depicted in Fig. 1 5.7(a). At one vertex we
have attached the current whose conservation we wish to study. Gauge
fields are attached to the other two . The anomaly that occurs typically has
the form
oµ Jµ ,...., c,µv p J.. Fµv Fp J.. .
Analogous phenomena can occur in higher dimensions, but then it is
necessary to consider diagrams with more external lines. The reason is
simply that the c, symbol has D indices in D dimensions. Thus in ten
dimensions, for example, the simplest anomaly has the structure
oµ Jµ "' eµ 1 µ2 · ' " µ • 0Fµ 1 µ • • • Fµ9µ 1 ·
2 0
The corresponding Feynman diagram must have at least six legs (hexagon
graph) as shown in Fig . 1 5 .7(b).
In the case of string theories the situation is similar, but one should
consider a Feynman diagram appropriate to string theory, i.e. a two
dimensional surface. For example, in the case of a theory involving closed
oriented strings the only relevant diagram is a torus diagram with external
strings (one to represent the gauge current and five to represent gauge fields),
as shown in Fig. 1 5.8. In the case of type IIA superstrings all anomalies
trivially cancel because of the left-right symmetry of the theory. In the case
of the type IIB theory and the heterotic string theories the spectrum is not
Fig. 15.7. Gauge-current anomalies in D = 4 can occur in triangle graphs

such as the one in (a). In D = 10 the simplest anomalous graph is the hexagon
diagram shown in (b).
A
(a)
J A
(b)
Fig. 15.8. The simplest potentially anomalous diagram in a D = 10 theory of

closed oriented strings is the torus with six external massless strings.
668 J. H. Schwarz
symmetric and the cancellation of anomalies is very non-trivial . In fact, the

full-fledged string calculations have not yet been carried out. However, there
is evidence (based on a low-energy expansion described below) that the
cancellation does in fact take place for the theories listed in Table 1 5 . 1 [ 1 1 ,
14, 30] . I n the case of type I superstrings surfaces of various different
topologies need to be studied . The two that are relevant to pure gauge
anomalies were evaluated by Green and me in 1984 [3 1] . We found that they
individually give anomalies but that the anomaly cancels from the sum for
the group choice S0(32) only.
At low energies, string theories can be approximated by point-particle
theories. This is physically reasonable since, when the wavelengths that
occur are much longer than the strings, the spatial extension of the string
becomes irrelevant. In practice one associates a quantum field with each
massless mode of the string theory and writes an effective Lagrangian in
terms of these fields only. The effects of massive string modes, and stringiness
in general , are expanded out in a series of terms involving higher powers of
derivatives and fields. The correction terms in such an expansion are
typically suppressed by powers of E/Mp , where Mp is about 10 1 9 GeV . Since
the anomaly cancellation must take place at each order in this expansion,
such a formulation is sufficient for investigating the cancellations in the
leading orders of the low-energy expansion. This al ready gives very severe
constraints. It is believed that if the leading-order constraints are satisfied
then all the higher ones will also be, but this is not yet completely
established.
For theories with N = 1 supersymmetry in ten dimensions the
contributions to hexagon anomalies arise entirely from chiral fermions.
There is one Majorana-Weyl gravitino that makes a certain contribution ,
an additional Majorana-Weyl spinor from the supergravity multiplet, and n
Majorana-Weyl spinors from the Yang-Mills supermultiplet for a gauge
group with n generators. By considering the pure gravitational anomaly
arising from hexagon diagrams with six external graviton lines one learns
that a necessary condition for anomaly cancellation is that the gauge group
have 496 generators. This requirement is satisfied , in particular, by S0(32)
which has 1 · 32 · 3 1 = 496 generators and by E s x E s which has 248 + 248 =
496 generators [ 1 1] . Thus one prediction of superstring theory is that there
are 484 new forces beyond the 12 that are al ready known . (This counting
does not include gravity, which is not associated with a Yang-Mills
generator.) Of course, if this is to be reconciled with nature, almost all these
symmetries must be broken at a very high mass scale.
There are many groups in addition to the two that have been mentioned
that have 496 generators. To obtain additional restrictions we consider
anomaly diagrams involving Yang-Mills fields as well as gravitons. The
cancellation of all such anomalies leads to the additional requirement that
an arbitrary generator F of the algebra, expressed in the adjoint
representation , should satisfy the equation [1 1]
Tr F6 = la Tr F 2 Tr F4 - 1
4loo {Tr F 2 )3•
Remarkably this equation, with exactly the right coefficients, is satisfied for
both S0(32) and Ea x Ea. The only other possible solutions are Ea x
[U(1)] 2 4 a and [U(l)J496, neither of which seems to correspond to a string
theory. In the anomaly analysis an antisymmetric tensor field Bµ v that is part
of the supergravity multiplet plays an important role. The details are given
in the references.
The SO( 16) x SO( 16) heterotic string theory, listed in Table 15. 1 , has only
240 generators. This is possible because it is not supersymmetric. It is chiral ,
however, and the cancellation of anomalies involves 'miracles' analogous to
those of the other theories [ 14] .
The groups S0(32) and E a x Ea , singled out by the anomaly cancellation
analysis in superstring theories, had previously arisen in another context
[32] . Specifically, mathematicians have investigated lattices, like those of
solid-state physics, in higher dimensions. They were led to consider in
particular lattices that are self-dual (i .e. are coincident with the dual lattice)
for which the distance squared of each lattice site from the origin is an even
integer. (A cubic lattice generated by orthogonal unit vectors is self-dual but
not even.) It turns out that even self-dual lattices only occur in dimensions
that are multiples of eight. In eight dimensions there is just one and it is
generated by the root vectors of the Lie algebra of Ea. In 16 dimensions there
are two of them . One is obvious from the eight-dimensional construction,
namely the root lattice generated by Ea x Ea· The second possibility is
closely associated with S0(32). It is generated by root vectors of the algebra
S0(32) and certain spinorial weights of its covering group spin(32) . More
precisely, it is the weight lattice of the group spin(32)/Z 2 • This coincidence
between self-dual lattices and the algeb ras singled out by the anomaly
analysis was skillfully exploited in the construction of the heterotic string
theories. The basic idea is that modes that travel clockwise (right-moving)
on the string are described by the mathematics of a ten-dimensional
superstring whereas those that travel counterclockwise (left-moving) are
described by the mathematics of a 26-dimensional bosonic string. Ten of the
670 J. H. Schwarz
26 are paired with the right-moving coordinates to describe 'ordinary' ten

dimensional spacetime, whereas the remaining 16 are required to form a 1 6-
dimensional torus that is conjugate to one of the two even self-dual lattices.
These degrees of freedom, which are the origin of the Yang-Mills symmetry
structure of the theory, can equivalently be described using fermionic
coordinates instead [ 1 3] .
1 5.6 Compactification
It may be that our theoretical understanding of superstring theory is not yet
sufficiently developed to be able to do correct phenomenology . If we
nevertheless choose to plunge in with reckless abandon, then it is clear that a
crucial question is what to do with the six extra spatial dimensions. The
natural guess is that they curl up to form a six-dimensional space K that is
sufficiently small not to have been observed . In fact , the only fundamental
scale in the theory is the Planck length, and it is natural to suppose that the
internal space K is roughly of this size. This means that it is about the same
size that is characteristic of strings themselves.
Since string theory contains gravity it should determine spacetime
geometry dynamically� and so we must require that the 'background
geometry' M 4 x K corresponds to a solution of the classical equations of
motion. That is the classical statement sometimes called 'spontaneous
compactification'. Quantum mechanically, it should be determined as part
of the characterization of the vacuum : the quantum state of lowest energy. It
is generally assumed , since that is all we can do at this point, that a classical
solution is a good approximation to a quantum ground state. This might not
be true. Another potential pitfall arises from describing K in terms of
classical differential geometry. In doing this it is implicit that that geometry
is determined entirely in terms of the gravitational field (metric tensor) .
However, this is just one of an infinity of modes of the string . Of course, it is
singled out by being massless. The massive modes could play an equally
important role, however, in characterizing the geometry and topology of K
if K is not much larger than the characteristic string length scale. In this case
a whole new type of geometry, let us call it 'string geometry', may be
required for a suitable description of K . Since this does not yet exist, one
assumes that it is not necessary. I would not be surprised , however, if the
prevailing opinion changes when these matters are better understood .
Whatever the correct language for describing it may turn out to be, once
the vacuum of the theory is identified correctly we shall be in a strong
position for calculating many quantities of physical interest. The particle
Superstring unification 67 1
spectrum would be determined by a small oscillation analysis, i.e. by

studying low-lying excitation modes. Also , as it turns out, many interesting
quantities are controlled by the topology of K. Thus one can go a long way
with qualitative topological information rather than quantitative geometric
information . For example, in a large class of models that has been
considered the number of generations of quarks and leptons is controlled by
the Euler characteristic of K
Ngen = 1 1 x (K )I .
A rather specific compactification scenario was proposed in the paper of
Candelas, Horowitz, Strominger and Witten [33] . They begin with the
Ea x Ea theory and argue that a classical solution is obtained if K is a
Calabi-Y au space, a Kahler manifold of S U(3) holonomy (vanishing first
Chern class). Such a manifold always admits a Ricci-flat metric. It has
recently been realized that, beginning at fourth order in an expansion in the
string length scale, this is not the metric that solves the string equations of
motion [34] . However, it has been argued convincingly that there is always
another 'nearby' metric that does [35] . In formulating the solution one
identifies the connection of the space K with an S U(3) subgroup of Ea x Ea .
In fact it is embedded entirely in one Ea factor which thereby breaks down to
E6 , from which the usual gauge group should emerge.
The E6 group described above can be broken further if K is not simply
connected . In this case there are non-contractible loops in K around which
gauge fields can wrap to give non-zero Wilson-loop integrals even though
the corresponding field strengths vanish. (This is analogous to the Bohm
Aharanov effect.) In this way it is possible to break E6 down to a group close
to the standard model but containing at least an extra U( 1). This would
imply the existence of a second Z boson at the electroweak scale. There are
several alternative scenarios, however.
One alternative 'superstring-inspired model' [36] starts with a specific
Calabi-Yau space K with x = 6 and fundamental group rr1 = Z 3 . Wilson
loops are used to break E6 to S U(3) x S U(3) x S U(3). Then , by introducing
some optimistic assumptions about a resulting potential, it is suggested that
this could break to S U(3) x S U(2) x U( 1) at an intermediate scale by a
standard Higgs mechanism. I am sure Ross and his collaborators would
agree that this is unlikely to be the final correct model . Still, it is remarkable
how close they can come to accounting for a targe number of desiderata in
this way. There are indications that the axions may not have the required
properties, however [37] .
672 J. H. Schwarz
15.7 Remaining problems and conclusions

Superstring theory is an enormously ambitious attempt to account for all
properties of fundamental forces and particles. It starts by describing the
physics at the Planck scale. From this all the physics at ordinary energies
should be derivable. However, the gap is so large that ordinary physics is, in
a sense, all hyperfine structure. Working out these connections in detail is
surely going to be a very long struggle requiring the efforts of many clever
people. Whether it will ultimately succeed is hard to foresee, even if we
assume that the theory is correct. The experimental information that will be
learned from the next generation of accelerators, especially as concerns
supersymmetry and the Higgs sector, should provide very valuable clues.
Without it we are unlikely to find our way.
There are a number of theoretical issues, less directly connected with
phenomenology, that must also be addressed. We would like to know how
many consistent superstring theories there are. Ideally, there is just one and
it explains everything, but there is certainly no guarantee that this is the case.
Further work is required in formulating the theories. It seems likely that a
deep and beautiful principle underlies string theory, but it remains to be
elucidated. The work on string field theory is one approach that is being
pursued. Some other approach may, however, be required . One alternative
that looks interesting is based on the analytic geometric of conformal field
theory [38] . Whatever the answer may be, it will serve to define the theory
non-perturbatively and undoubtedly lead to new insights. It could happen
that by the time the dust settles the subject will look radically different than
it does today.
Superstring theory was developed primarily because it seemed promising
for overcoming the divergence problems of quantum gravity. However, even
classical general relativity has its problems. It has been proved that generic
initial data lead to singularities. This is bad because the structure of
spacetime subsequent to the formation of the singularity is not determined .
The theory is therefore incomplete. It is a plausible conj ecture that classical
superstring theory is not subject to the same problems (these remarks are
mostly due to D . Gross) . They are short-distance effects , and it is at short
distances that the theory is modified. The 1986 result [34] that a general
solution of Rµ v = 0 is not a solution of string theory is encouraging , since it
implies that the Schwarzschild solution, with its singularity, is averted . The
actual proof that string theory is a complete classical theory will
undoub tedly require a lot of work. The discovery of the optimal formulation
and underlying principles discussed above is probably a necessary

preliminary.
Other questions that need to be studied are why a particular ground state
of the form M 4 x K should be singled out, if that is the case. It would be very
sad if there were thousands of theoretically acceptable vacua, since then
much predictivity would be lost despite the uniqueness of the underlying
theory. We need to understand the origin of the mass hierarchy mw /mp ,..,.,
10 - 1 7 and the origin of the supersymmetry breaking. These are tough
problems , but the one that really worries us the most is to understand why
the cosmological constant IAI is less than 10 - 1 2 0 in natural units. Every little
effect one can think of makes a much larger contribution to A than this. Why
they should all precisely cancel is, at the moment, totally baffling.
In conclusion , the theoretical understanding of superstrings is
progressing rapidly. The long-term prospects for the subject appear very
bright. Developing this subject is an exhilarating intellectual experience, but
it requires a major commitment of time and effort that is not appropriate for
everyone. In any case, there is plenty of work to keep busy those who choose
to make this commitment while we await the next round of experimental
results.
References
1 . Schwarz, J. H . ( 1985). Superstrings. The First Fifteen Years of Superstring Theory.

World Scientific ;
Green , M . B . , Schwarz, J . H . and Witten, E. ( 1 987). Superstring Theory. Cambridge
2. Goroff, M. and Sagnotti, A. ( 1986) . Nucl. Phys., 8266, 709.
3. Jacob, M . (ed .) ( 1 974) . Dual Theory. Physics Reports Reprint Volume I. North
Holland : Amsterdam;
Scherk, J . ( 1975). An introduction to the theory of dual models and strings. Rev.
Mod. Phys., 47, 123.
4. Neveu, A . and Schwarz, J. H. ( 197 1). Nuc/. Phys., 831 , 86; Phys. Rev., D4, 1 108 ;
Neveu, A . , Schwarz, J . H . and Thorn, C. B. ( 197 1). Phys. Lett., 358, 529;
Thorn, C. B. ( 197 1). Phys. Rev., D4, 1 1 12 .
5. Ramond, P . ( 197 1). Phys. Rev., 03, 24 15.
6. Gliozzi, F., Scherk, J. and Olive, D. ( 1976). Phys. Lett., 658, 282; (1 977) Nucl. Phys.,
B l22, 253.
7. Green, M . B. and Schwarz, J. H . (198 1). Nucl. Phys., 8181 , 502.
8. Goddard, P. and Thorn, C. B. ( 1972). Phys. Lett., 408, 235 ;
Schwarz, J. H. ( 1972). Nuc/. Phys., 846, 6 1 ;
Brower, R . C. and Friedman , K . A . ( 1 973). Phys. Rev., D 1 , 535.
9. Scherk, J. and Schwarz, J. H. ( 1974). Nucl. Phys., 881 , 1 1 8.
674 J. H. Schwarz
10. Yoneya, T. ( 1974). Prog . Theor. Phys., 51, 1907.

1 1. Green, M. B. and Schwarz, J. H. ( 1984). Phys. Lett., 149B, 1 17 .
12. Green, M . B . and Schwarz, J . H . ( 1 982). Phys. Lett., 109B, 444.
13. Gross, D. J., Harvey, J. A . , M artinec, E. and Rohm, R. ( 1985). Phys. Rev. Lett ., 54 ,
502 ; ( 1 985) Nucl. Phys., B256, 253 ; ( 1 986) Nucl. Phys., B267, 7 5 .
14. Dixon, L. and Harvey, J. A. ( 1986). Nucl. Phys., B274, 9 3 ;
Alvarez-Gaume, L . , Ginsparg, P . , Moore, G. and Vafa, C. ( 1986). Phys. Lett., 1 7 1 B ,
155.
15. Seiberg, N . and Witten, E. ( 1986). Nucl. Phys., B276, 272 .
16. Brink, L., Di Vecchia , P . and Howe, P. ( 1976). Phys. Lett., 65B, 47 1 ;
Deser, S . and Zumino , B . ( 1976). Phys. Lett., 65B, 369.
17. Polyakov, A . M . ( 198 1 ) . Phys. Lett., 103B, 207, 2 1 1 .
18. Alvarez, 0. ( 1983). Nucl. Phys., B2 16, 125;
D'Hoker, E. and Phong, D. ( 1 986). Nucl. Phys., B269, 205 ;
Belavin, A. and Knizhnik, V. ( 1 986). Phys. Lett., 168B, 20 1 .
Alvarez-Gaume, L . , Moore, G. and Vafa, C. ( 1 986). Commun. Math. Phys., 106, 1 ;
Mannin, Yu . I . ( 1 986). Pisma ZETP, 43, 1 6 1 ; ( 1 986) Phys. Lett., 1 72B, 184;
Bost, J. B . and Nelson, P. Harvard preprint HUTP-86-A0 14;
Bost, J . B . and Jolicoeur, T. ( 1 986). Phys. Lett., 1 74B, 273 ;
Catenacci, R . , Cornalba, M ., Martellini , M . and Reina , C. ( 1 986). Phys. Lett. , 1 72B,
328 ;
Gomez, C. ( 1 986). Phys. Lett., 175B, 32;
Alvarez-Gaume, L., .\'Vloore, G., Nelson, P., Vafa, C. and Bost , J . B. H arvard
preprint HUTP-86/A039.
Moore, G., Harris, J . , Nelson, P. and Singer, I . Harvard preprint H UTP-86/A05 1 ;
Morozov, A . Preprint !TEP 86-88;
Belavin, A., Knizhnik, V . , Morozov, A. and Perelomov. Preprint !TEP 86-59 ;
Beilinson, A. A. and Manin , Yu . I . ( 1 986). Moscow preprints.
Restuccia, A. and Taylor, J. G . ( 1 986). Phys. Lett., 174B, 56.
19. Cremmer, E. and Gervais, J. L. ( 1974). Nucl. Phys., B76, 209; ( 1975) Nucl. Phys.,
B90, 4 10;
Kaku , M . and Kikkawa, K . ( 1 974). Phys. Rev., D lO, 1 1 10, 1 82 3 .
20. Green , M . B. and Schwarz, J . H . ( 1984). Nucl. Phys., B218, 43; ( 1 984) Nucl. Phys.,
B243, 475;
Green, M . B., Schwarz, J . H . and Brink, L. ( 1983) . Nucl. Phys., B219, 437 .
21. Siegel, W. ( 1985). Phys. Lett., 1 51B, 39 1 , 396;
Siegel, W . and Zwiebach, B . ( 1986) . Nucl. Phys., B263, 105 ;
22. Banks, T. and Peskin, M . ( 1986) . Nucl. Phys., B264, 5 13 ;
Peskin, M . E . and Thorn, C . B . ( 1 986) . Nucl. Phys., B269, 509.
23. Neveu, A. and West , P. C. ( 1985). Phys. Lett., 165B, 63, and CERN preprints.
24. Hata, H . , ltoh, K . , Kugo , T., Kunitomo, H . and Ogawa, K. ( 1986) . Phys. Lett.,
1 72B, 186, 195, and Kyoto University prepri nts.
25. Witten , E. ( 1 986) . Nucl. Phys., B268, 253 ; B276, 29 1 .
26. Giddings, S . B . , Martinec , E . and Witten , E . ( 1 986). Phys. Lett., 176B, 362.
27. Kato, M. and Ogawa, K. ( 1983). Nucl. Phys., B212, 443 ;
Hwang, S . ( 1983). Phys. Rev., 028, 26 14.
28 . Gross, D. and Jevicki, A. ( 1986). Princeton preprint ;
Cremmer, E., Schwimmer, A. and Thorn, C. B. Ecole N ormale Superieure preprint
LPTENS-86- 14.
29. Giddings, S. B. ( 1 986). Nucl. Phys., 8278, 242 ;

Giddings , S . B . and M artinec, E. ( 1986). Nucl. Phys., 8278, 9 1 .
30. Alvarez-Gaurne, L. and Witten, E . ( 1983). Nucl. Phys., 8234 , 269 .
3 1 . Green , M . B . and Schwarz, J . H . ( 1985). Nucl. Phys., 8255, 93.
32. Goddard, P. and Olive, D. ( 1985). In Vertex Operators in Mathematics and Physics,
ed . J . Lepowsky et al. , p. 5 1 . Springer : Berlin.
33. Candelas, P., Horowitz, G., Strorninger, A. and Witten, E. ( 1985). Nucl. Phys., 8258,
75.
34. Grisaru , M . T . , V a n de Van, A . E . M . and Zanon , D. ( 1986). Phys. Lett., 1 738, 423 ;
Freeman, M . D . and Pole, C. N . ( 1 986). Phys. Lett., 1 748, 48 ;
Gross, D . and Witten , E. ( 1986). Nucl. Phys., 8277, 1 .
35. Witten, E. ( 1986). Nucl. Phys., 8268, 79.
Nerneschansky, D. and Sen , A . ( 1986). SLAC preprint.
36. Greene, B. R . , Kirklin , K. H . , Miron , P. J. and Ross, G. G. ( 1986). Oxford
University preprints .
37 . Kirn, J . E. ( 1986) . Seoul National University preprint ;
Dine, M . and Seiberg, N. ( 1 986). Nucl. Phys., 8273, 109.
38. Friedan, D. and Shenker, S. ( 1987). Nucl. Phys., 828 1 , 509 .
16
Covariant description of canonical formalism
in g eometrical theories
C ED O M I R C RN K O V I C and E D W A R D W I T T EN t
Quantum field theories are usually studied either by means of path integrals
or by means of canonical quantization. Path integral quantization has the
great virtue of explicitly maintaining all relevant symmetries, such as
Poincare invariance. The canonical approach is usually interpreted as an
approach that ruins Poincare invariance from the beginning through an
explicit choice of a 'time' coordinate. This is not necessarily so, however.
The essence of the canonical formalism can be developed in a way that
manifestly preserves all relevant symmetries, including Poincare invariance
(Witten , 1986; Zuckerman , 1986). The purpose of the present paper is to
carry this out in the case of non-abelian gauge theories and general
relativity.
In the canonical formalism of a theory with N degrees of freedom, one
usually introduces coordinates and momenta p i and qi , i,j = 1 , . . . , N. One
then defines the two-form
(1)
1
It is convenient to combine the p i and qi i n a variable Q , I = 1 , . . . , 2N, with
Q i = p i for i � N and Q i = q i - N for i > N. One can think of w as an
antisymmetric 2N x 2N matrix W u whose non-zero matrix elements are
w i , i + N = - w i + N, i = 1 . This matrix is invertible; we will denote the inverse
matrix as wu. One defines the Poisson bracket of any two functions A(Q 1 )
and B(Q 1 ) by
[ an
u oA
A , BJ = w a 1 a r
Q Q (2)
t Research supported in part by the National Science Foundation under grants P HY80-
19754 and PHY86- 16 129.
Geometrical theories 677
Given the form of ( 1 ), this is easily seen to coincide with the usual definitions
of Poisson brackets. The advantage of the definition in (2) is that, as is well
known (see e.g . Abraham and Marsden , 1967), the essential features of w can
be described in an invariant way. Let Z be the phase space of the theory
under discussion , that is , the space on which the ps and qs are coordinates. If
one interprets w as a two-form on Z, then it is clearly a closed two-form,
dw = O, (3)
since its components are constant in the coordinate system used in ( 1). What
is more, we have already noted that the matrix w is invertible. The converse
to this is as follows. Let w be any two-form on a manifold Z (which for us will
be the phase space of a physical theory). Suppose that w is closed (obeys (3)),
and is non-degenerate in the sense that at each point z E Z , the matrix w ii (z)
is invertible. Then it is a classical theorem that locally one can introduce
coordinates on Z to put w in the standard form ( 1 ). (This is not true globally
in theories with interesting geometrical content.) A non-degenerate closed
two-form is called a symplectic structure. Thus, to describe the canonical
formalism of a theory it is not at all necessary to find or choose ps and qs; the
essence of the matter is to describe a symplectic structure on the classical
phase space.
Clearly, the notion of a 'symplectic structure on phase space' is a more
intrinsic concept than' the idea of choosing ps and qs. However, at first sight
it might appear that the very concept of phase space is a non-covariant
concept, tied to a non-covariant , Hamiltonian description . This is not really
so . The whole idea , classically, of picking ps and qs is that the initial values of
the ps and qs determine a solution of the classical equations. More precisely,
classical solutions of any given physical theory, in any given coordinate
system , are in one-to-one correspondence with the values of the ps and qs at
time zero . This simple consideration leads us to a manifestly covariant
definition of what we mean by classical phase space : in a given physical
theory, classical phase space is the space of solutions of the classical
equations. We can always, if we wish, pick a coordinate system and identify
the classical solutions with the initial data in that coordinate system, but
there is no necessity to make such a non-covariant choice.
Given a relativistic field theory, such as the scalar field theory with
Lagrangian
L= LG O,</> 0'</> - V(</>)) , (4)
(here M is spacetime, and </> is a scalar field), how do we go about finding a

67 8 C. Crnkovic and E. Witten
covariant description of a symplectic structure on the space, Z, of classical

solutions? A point in Z is a solution of the classical equations
0 = �¢ - V'(</J) , (5)
with � = - orx arx being the standard Laplacian. In order to construct a
symplectic structure on Z , we will need to discuss functions, tangent vectors,
differential forms, and exterior derivatives on Z.
Functions on Z are, of course, the easiest to discuss. Among the most
important are the following . Let x E M be a spacetime point. If ¢ is a solution
of (5) , its value ¢(x) at the point x is a real number. The mapping from the
function ¢ to the number </J(x) is a real valued function on Z. We will denote
this function by ¢(x).
Now we will consider tangent vectors. At a point p E Z, corresponding to a
solution of (5), a tangent vector would be a small displacement in ¢ which
preserves (5). Thus, a tangent vector is the same as a solution of the
linearized equations which we obtain by expanding around a solution of (5).
Requiring that ¢ = ¢ + b¢ should obey (5) to lowest order in b¢, we find
0 = � b¢ - V"(</J) b</J. (6)
A solution of (6) is a tangent vector at the point in phase space
corresponding to the solution ¢ . The tangent space is the vector space T of
solutions of (6).
Now, how do we describe one-forms on Z? The space of one-forms is, of
course, the dual of the tangent space which we have just described ; it is the
space of linear functionals on T. Most important for our purposes are certain
one-forms on Z which we will now describe. Let x E M be a spacetime point.
For every solution b¢ of (6), its value b¢(x) at the point x is, of course, a
number. The transformation from the function b¢ to the number b¢(x) is a
one-form on phase space which we will call b¢(x) . More generally, we can
make k-forms as wedge functions of the one-forms b¢(x) :
I
A = dx 1 . . • dx. a.,. . . ( </> ) b <f> ( x 1 ) b¢(x 2 ) . . . b <f> (x. ).
. x, (7)
Here, for each n-tu ple of spacetime points x 1 . . . x, p r.xx , . . . x. ( ¢) is an

arbitrary zero form - an arbitrary real valued function on Z. On the right
hand side of (7), the product of the b¢ (xJ is understood to be a wedge
product. In particular, as the wedge product of one-forms is anticommuting ,
we interpret the b¢(x) as anticommuting objects :
b¢(x) b</J( y) = - b</J( y) b¢(x). (8)
The general n-form on Z can be expanded as in (7) ; this expansion , however,
Geometrical theories 67 9
is not unique, since the b</J(x) are not linearly independent, being subject
to (6).
Finally, we need an exterior derivative on Z, which we will call b and
which must map k forms to (k + 1)-forms. It should obey
b2 = 0 (9)
and the Leibniz rule
b(AB) = bA B + ( - l)AA · bB.
· ( 1 0)
We define b by saying that acting on the zero-form (function) </J(x) , b gives
the one-form that we have called b<fJ(x) . The Leibniz rule then determines the
f
action of b on an arbitrary zero-form r :
br
b(r( ¢ )) = dx b</J(x), ( 1 1)
b</J(x)
where br/b</J(x) is just the variational derivative of r with respect to </J(x).
Equation ( 1 1) is a familiar formula to which we are giving, perhaps, a
slightly novel meaning . To act with b on k-forms of k > 0, one must bear in
mind, first of all , that as the one-form b</J(x) is the exterior derivative of the
zero-form </J(x), it must be closed :
b(b</J(x)) = 0. ( 12)
Also , using the Leibniz rule, we then have the exterior derivative of a general
k form (7) :
f
bA � dx 0 . . . dx, foJ�(;:\ ¢)
b ¢ (x0) b ¢ (x 1 ) . . . J ¢ (x, ). ( 13)
Although our definition of b has been rather formal , one can readily see that
it possesses the standard properties of the exterior derivative. Thus, if V is a
vector field, A a zero-form, and iv the operation of contraction with V, we
have V(A) = iv bA .
Having defined the relevant concepts, how are we to find a symplectic
structure in , say, the scalar field theory (4)? The idea (Witten, 1986 ;
Zuckerman , 1986) is to consider the 'symplectic current'
( 14)
At each spacetime point, ( 14) is a two-form on Z ; but in its dependence on x,
la is a conserved current :
( 15)
To verify ( 1 5) , one needs the equation of motion (6) and the fact that b ¢ is
anticommuting. As la is conserved, its integral over an initial value
680 C. Crnkovic and E. Witten
hypersurface L,
( 16)
is independent of the choice of L and so in particular is Poincare invariant.

The two-form w is our desired symplectic structure on Z. It is evidently
closed, in view of ( 12), and it is easy to see that upon picking I: to be the
standard initial value surface t = O, ( 16) reduces to
w= L b</> bef,, ( 17)
which (as in ( 1)) is the standa.rd formula .

Our goal in the present paper is to implement this procedure in the case of
Yang-Mills theory and general relativity. The main novelty that arises is the
need to establish gauge invariance as well as Poincare invariance of the
symplectic structure. We will consider the two cases in turn .
16.1 Yang_:Mills theory

Considering first Yang-Mills theory, let A be the gauge connection and F
the Yang-Milk curvature or field strength. The covariant derivative of a
charged field A is Dal\ = aa/\ + [Aa, A] . The Yang-Mills equation of motion
lS
0 = [D µ , Fµ v] = aµFµ " + [Aµ , Fµ v] . ( 18)

The variation of ( 18) is
Dµ bFµ v + [ bAµ ' Fµ v] = 0. ( 1 9)
We define the symplectic current Ja(x) as
ix J = Tr[bAµ bFµix] , (20)
where
bFµ11. = Dµ bAa - D11. bAµ . (2 1 )
Let us show that J11. is conserved . We have
a a
a11.Jix = Tr[D bAµ bF11.µ] + Tr[bAµD bF11.µ]
= ! Tr [ bF1µ bF11.µ] - Tr[bAµ [bAix, bFaµ]] = 0. (22)
On the last line of (22), the second term vanishes because bA is
anticommuting , and the first because bF is anticommuting . So the two-form
w= L Tr[bA" bF,,J dl:" (23 )
is Poincare invariant.
Geometrical theories 68 1
µ
It is easy to see that w is also closed ; 6A , being the exterior derivative of
µ
the zero-form A , is closed, while, in view of (2 1) and the anticommutativity
of b, we have
6(6FwJ = b(Dµ 6Aa - Da 6Aµ) = 6([Aµ, bAaJ - [Aa, bAµ] )
= { bAµ , bAa } - { 6Aa, bAµ } = 0 . (24)
Therefore, w is closed .
It remains to discuss the behavior of w under gauge transformations. First
of all , the gauge transformation law for the gauge field is
Aµ � Aµ + [Dµ, s] = Aµ + aµs + [Aµ, s] . (25)
Varying (25), we find that under gauge transformations, bA transforms as
bAµ � bAµ + [ JAµ, sJ . (26 )
In particular, JA transforms homogeneously under gauge transformations.
And JF transforms in the same way :
JFµa � JFµa + [JFµa ' s] . (27)
Consequently, r and w are gauge invariant.
This is an important step in the right direction, but it is not the end of the
story. Let i be the space of solutions of ( 18), and let Z be the space of
solutions of ( 18) modulo gauge transformations. Thus, Z = Z/G , with G
being the group of gauge transformations. So far we have defined a gauge
invariant closed two-form w on i. What we want is a gauge-invariant closed
two-form on Z . We would like to show that the differential form w that we
have defined on i is the pullback from Z to i of a differential form on Z,
which we will also call w . This will be so if the following condition is obeyed.
If V is any vector field tangent to the G orbits on i, and i v is the operation of
contraction with V, the requirement is ivw = 0. This is a fancy way of saying
that the components of w in the gauge directions are zero ; one must require
this since the gauge directions a re eliminated in passing from i to Z, so a
differential form on Z cannot have non-zero components in those directions.
In Yang-Mills theory, the gauge directions in field space are simply
JAµ = D µs. (28)
More generally, we can consider a field variation
JAµ = J' Aµ + Dµs, (29)
which has a gauge component Dµs, and another component J'Aµ which is
not pure gauge. To verify that w has vanishing components in the gauge
di rections, we must show that if we insert (29) in the definition of w, the term
proportional to s drops out. The expression which must vanish is
I I
�w = d L. Tr [D• • bF'• + bA. [F"" ' 6]] = dL. a Tr(ef"" ) .
µ
(30)
In the last step, we have used anticommutativity of s and bF11.µ and the
equation of motion ( 19) . Indeed , (30) vanishes, being the integral of a total
derivative, and this completes the construction of a symplectic structure on
the gauge-invariant space Z.
16.2 General relativity

We now turn our attention to general relativity. The discussion is similar,
although somewhat more complicated. Let g v be the metric tensor of
µ
spacetime, and R v11.P the corresponding Riemann tensor. The equation of
µ
motion of pure general relativity is the Einstein equation R v = 0. To find its
µ
variation, remember that for any vector field V, one has
[D , D J V11. = R11.;. v V;. . (3 1)
µ µ
One finds for the variation of the Einstein equation
(32)
Here
(33)
is the variation of the Levi-Civita connection r . Note that b r transforms as
a tensor. We define the symplectic current
111. = b r�v [bg µ + !gµ [) In gJ - [)r;v [bg 11.µ + !g11.µ [) In g J ,
v v
(34)
where [) In g = [) In det(g v ) = gµ v bg v = - g v bgµ v . Also keep in mind that
v 11. vP [) · µ µ µ
[)gµ = - gµ g g11.p
We have to show that D J11. = 0. Terms with b i n g give
v 11.
v
! (D )gµ [) In g - !(D
11. [)r�v )g11.µ
11. [)r;v gµ
[) In g + 1 [)r�v D11. [) In g
v
- 1 [)r;v g11.µ D 11. [) In g = [)r�v gµ {Jr;;. , (3 5 )
where we have used ( 33), (32) , the fact that the metric tensor is covariantly
constant, and the anticommutativity of b r . The remaining terms are
v v
(D(f. b r:v ) [)gµ + [)r:vv11. bgµ - (D(f. [)r;J [)g(f.1' - [)r;vD bg11.µ
11.
v
= - [)r vD [)gaµ = [) rv v [)rr1.µp g11.P '
11.
µ µ
which, together with (35) , gives D11.J11. = 0. This makes
W= i dL, jg J"
Poincare invariant.
Geometrical theories 683
Let us show that w is closed, 6w = 0. We have
bw = J. dE, (b jg r + Jg M')
c5r = - ! c5r� 6g µ" b I n g + ! br� ·[Jga µ c5 In g .
" "
Remembering that 6 In g is an anticommuting one-form whose square is
zero , we have
c)J<I = -!Ja b ln g .
As b Jg = ! Jg b ln g , 6w = O. Thus, w is closed .
It now remains to investigate gauge invariance of w . The fact that w is
invariant under diffeomorphisms is relatively trivial ; it follows from the fact
that all ingredients in the definition of w, including c5r, transform
homogeneously, like tensors. As in the Yang-Mills case, the more delicate
point is to show that we obtain a closed two-form not just on the space i of
solutions of the Einstein equations, but also on the subtler space Z = Z/G,
with G being the group of diffeomorphisms. We must show that components
of w tangent to the G orbits vanish. Under a diffeomorphism xµ -+ xu + eµ ,
the metric changes by
9µ v -+ 9µ v + Dµev + D veµ ·
W e assume e has compact support or, more generally, i s asymptotic at
infinity to a Killing vector field. It we write
bgµv = b 'gµ v + Dµ ev + Dv eµ , (36)
where the De terms are pure gauge but Sg is not, then our task is to show
that, with (36) inserted in w , the De terms do not contribute. It is useful first
to rewrite Ja as
r = !gaP (Dµ bg vp + D 6gµp - Dp bgµJ bgµ"
"
!
- Dµ bg µ a 6 In g + -! bg µaDµ b In g - !(Da b In g) b In g . (37)
Inserting (36) in (37), the term linear in e is (after dropping the ' from {J'g)
µ
/).. Jrl- = [Dµ , D a] e bg µ" + [D v , Dµ] bga e"
" µ
+ Dv [(Dµ bg " + D" b In g)ea + D" bgµaeµ
+ Dµea bgµ " + !D"ea b In g - (a � v )] , (38)
where we have used (3 3), (32), and the identity [Dµ , D ]e µ = O, which follows
"
from Rµv = O. We are entitled to discard from /),,Jµ terms of the form D Xµa,
a
with xµa being an antisymmetric tensor; such terms will vanish when
inserted in the integral for w. Discarding such total derivatives from (38), we
are left with

Rv A.µ a�r�· bg µ" + (R µA. vµ [Jg A.a + R aA. vµ bgµA. )e "
= (Rv A.µ a + R avµA. )eA. [)gµ v = gap Rp A.µ v£ A. [)g µ v = 0 ·
Therefore, components of w tangent to the action of the diffeomorphism

group are zero.
In conclusion, we have described in a manifestly covariant way the
foundations of the canonical formalism of Yang-Mills theory and general
relativity. Since a similar treatmeqt of string theory has been given
elsewhere, it seems that such an approach is possible for all of the
geometrical theories in physics.
References
Abraham, R. and Marsden, J. E. ( 1 967). Foundations of Mechanics. Benjamin : New

York.
Witten, E. ( 1 986) . Interacting field theory of open superstrings. Nucl. Phys., B276, 29 1 .
Zuckerman , G. ( 1 986). Action functionals and global geometry. Yale University preprint,
to appear in the- proceedings of the San Diego workshop, ed. S .-T. Yau et al.

Three Hundred Years of Gravitation - Stephen Hawking and Werner Israel (Eds.) - 2022

Uploaded by

Copyright:

Available Formats

Three Hundred Years of Gravitation - Stephen Hawking and Werner Israel (Eds.) - 2022

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Three Hundred Years of Gravitation - Stephen Hawking and Werner Israel (Eds.) - 2022

Uploaded by

Copyright:

Available Formats

Autore JS. NE Tl'T 0 N, 1rin. Coll. Cantab. Soc. Mathefeos

Th« right tJ/tltt!

CAMBRIDGE UNIVERSITY PRESS

New York New Rochelle Melbourne Sydney

© Cambridge University Press 1987

First published 1 987

Printed and bound in Great Britain by

British Library cataloguing in publication data

Library of Congress cataloguing in publication data

ISBN 0 521 34312 7

4.2 Theoretical framework 52

4.3 The inverse square law 56

4.4 The weak equivalence principle 60

4.5 Preferred frames and locations 65

4.6 Additional deviations from general relativity 67

4.7 The measurement of the constant of gravitation 70

S Experimental gravitation from Newton's Principia to Einstein's

5.2 Newtonian gravitation and general relativity: commonalities and contrasts 82

5.4 Testing the applications of gravitation theory 101

5.5 Beyond the tercentenary 120

6 The problem of motion in Newtonian and Einsteinian gravity 128

8.3 Observational evidence for black holes 297

10 The emergence of structure in the universe: galaxy formation and

11 Gravitational interactions of cosmic strings 499

1 1.14 Gravitational field of waves on infinite strings 521

12 Inflationary cosmology 524

S. K. Blau and A. H. Guth

12.6 Problems of the original inflationary model 550

12.7 The new inflationary universe 555

12.10 Inflation in the minimal SU(5) grand unified theory 579

12.11 False vacuum bubbles and child universes 59 0

12.12 Conclusion 594

13 Inflation and quantum cosmology 604

16 Covariant description of canonical formalism in geometrical

On 5 July 1687 Edmond Halley wrote from London to Isaac Newton in

t General Relativity: An Einstein Centenary Survey, ed . S . W. Hawking and W . Israel.

which galaxies form. The mechanism of galaxy formation is being actively

The Philosophiae Natura/is Principia Mathematica b y Isaac Newton, first

magnitude and direction. Newton seems to have solved the problem

The written version o f my talk at the Newton tercentenary celebration in

Newton was a complicated man , who lived in complicated times. It would

there ou rselves. I think nothing was more dramatic as an example of our

purposes incapable of calculating what will happen next in the rotation of

through to a quantitative understanding of nature i n all its aspects. And

gravitational radiation, the so-called graviton, and yet it was supposed to be

3. 1 Newton's corpuscular-undulatory view of light

this apparent anticipation of modern quantum theory - a theory which was

3.2 Stated reasons for rejection of wave theory

discovered by Thomas Young a century later, an effective rectilinear

secondary nature. If new particles are to be invoked in order to compose

3.3 Newton and relativity

not a necessary principle. It might even be that the technicalities involved in

3.4 Newton's route to an undulatory-corpuscular picture?

the transmutation of substances, must have convinced him of a deep

3.5 Quantum mechanics

3.6 Physical reality

3.7 Reality of the state vector

that if measured in any one of these n directions, then with certainty it